Daily Papers

Computer use agents are LLM-based agents that can directly interact with a graphical user interface, by processing screenshots or accessibility trees. While these systems are gaining popularity, their safety has been largely overlooked, despite the fact that evaluating and understanding their potential for harmful behavior is essential for widespread adoption. To address this gap, we introduce OS-Harm, a new benchmark for measuring safety of computer use agents. OS-Harm is built on top of the OSWorld environment and aims to test models across three categories of harm: deliberate user misuse, prompt injection attacks, and model misbehavior. To cover these cases, we create 150 tasks that span several types of safety violations (harassment, copyright infringement, disinformation, data exfiltration, etc.) and require the agent to interact with a variety of OS applications (email client, code editor, browser, etc.). Moreover, we propose an automated judge to evaluate both accuracy and safety of agents that achieves high agreement with human annotations (0.76 and 0.79 F1 score). We evaluate computer use agents based on a range of frontier models - such as o4-mini, Claude 3.7 Sonnet, Gemini 2.5 Pro - and provide insights into their safety. In particular, all models tend to directly comply with many deliberate misuse queries, are relatively vulnerable to static prompt injections, and occasionally perform unsafe actions. The OS-Harm benchmark is available at https://github.com/tml-epfl/os-harm.

Mitigating reward hacking--where AI systems misbehave due to flaws or misspecifications in their learning objectives--remains a key challenge in constructing capable and aligned models. We show that we can monitor a frontier reasoning model, such as OpenAI o3-mini, for reward hacking in agentic coding environments by using another LLM that observes the model's chain-of-thought (CoT) reasoning. CoT monitoring can be far more effective than monitoring agent actions and outputs alone, and we further found that a LLM weaker than o3-mini, namely GPT-4o, can effectively monitor a stronger model. Because CoT monitors can be effective at detecting exploits, it is natural to ask whether those exploits can be suppressed by incorporating a CoT monitor directly into the agent's training objective. While we show that integrating CoT monitors into the reinforcement learning reward can indeed produce more capable and more aligned agents in the low optimization regime, we find that with too much optimization, agents learn obfuscated reward hacking, hiding their intent within the CoT while still exhibiting a significant rate of reward hacking. Because it is difficult to tell when CoTs have become obfuscated, it may be necessary to pay a monitorability tax by not applying strong optimization pressures directly to the chain-of-thought, ensuring that CoTs remain monitorable and useful for detecting misaligned behavior.

As AI becomes more capable, we entrust it with more general and consequential tasks. The risks from failure grow more severe with increasing task scope. It is therefore important to understand how extremely capable AI models will fail: Will they fail by systematically pursuing goals we do not intend? Or will they fail by being a hot mess, and taking nonsensical actions that do not further any goal? We operationalize this question using a bias-variance decomposition of the errors made by AI models: An AI's incoherence on a task is measured over test-time randomness as the fraction of its error that stems from variance rather than bias in task outcome. Across all tasks and frontier models we measure, the longer models spend reasoning and taking actions, the more incoherent their failures become. Incoherence changes with model scale in a way that is experiment dependent. However, in several settings, larger, more capable models are more incoherent than smaller models. Consequently, scale alone seems unlikely to eliminate incoherence. Instead, as more capable AIs pursue harder tasks, requiring more sequential action and thought, our results predict failures to be accompanied by more incoherent behavior. This suggests a future where AIs sometimes cause industrial accidents (due to unpredictable misbehavior), but are less likely to exhibit consistent pursuit of a misaligned goal. This increases the relative importance of alignment research targeting reward hacking or goal misspecification.

Large Language Models (LLMs) are frequently updated due to data or architecture changes to improve their performance. When updating models, developers often focus on increasing overall performance metrics with less emphasis on being compatible with previous model versions. However, users often build a mental model of the functionality and capabilities of a particular machine learning model they are interacting with. They have to adapt their mental model with every update -- a draining task that can lead to user dissatisfaction. In practice, fine-tuned downstream task adapters rely on pretrained LLM base models. When these base models are updated, these user-facing downstream task models experience instance regression or negative flips -- previously correct instances are now predicted incorrectly. This happens even when the downstream task training procedures remain identical. Our work aims to provide seamless model updates to a user in two ways. First, we provide evaluation metrics for a notion of compatibility to prior model versions, specifically for generative tasks but also applicable for discriminative tasks. We observe regression and inconsistencies between different model versions on a diverse set of tasks and model updates. Second, we propose a training strategy to minimize the number of inconsistencies in model updates, involving training of a compatibility model that can enhance task fine-tuned language models. We reduce negative flips -- instances where a prior model version was correct, but a new model incorrect -- by up to 40% from Llama 1 to Llama 2.

We demonstrate a situation in which Large Language Models, trained to be helpful, harmless, and honest, can display misaligned behavior and strategically deceive their users about this behavior without being instructed to do so. Concretely, we deploy GPT-4 as an agent in a realistic, simulated environment, where it assumes the role of an autonomous stock trading agent. Within this environment, the model obtains an insider tip about a lucrative stock trade and acts upon it despite knowing that insider trading is disapproved of by company management. When reporting to its manager, the model consistently hides the genuine reasons behind its trading decision. We perform a brief investigation of how this behavior varies under changes to the setting, such as removing model access to a reasoning scratchpad, attempting to prevent the misaligned behavior by changing system instructions, changing the amount of pressure the model is under, varying the perceived risk of getting caught, and making other simple changes to the environment. To our knowledge, this is the first demonstration of Large Language Models trained to be helpful, harmless, and honest, strategically deceiving their users in a realistic situation without direct instructions or training for deception.

Large language model (LLM) developers aim for their models to be honest, helpful, and harmless. However, when faced with malicious requests, models are trained to refuse, sacrificing helpfulness. We show that frontier LLMs can develop a preference for dishonesty as a new strategy, even when other options are available. Affected models respond to harmful requests with outputs that sound harmful but are subtly incorrect or otherwise harmless in practice. This behavior emerges with hard-to-predict variations even within models from the same model family. We find no apparent cause for the propensity to deceive, but we show that more capable models are better at executing this strategy. Strategic dishonesty already has a practical impact on safety evaluations, as we show that dishonest responses fool all output-based monitors used to detect jailbreaks that we test, rendering benchmark scores unreliable. Further, strategic dishonesty can act like a honeypot against malicious users, which noticeably obfuscates prior jailbreak attacks. While output monitors fail, we show that linear probes on internal activations can be used to reliably detect strategic dishonesty. We validate probes on datasets with verifiable outcomes and by using their features as steering vectors. Overall, we consider strategic dishonesty as a concrete example of a broader concern that alignment of LLMs is hard to control, especially when helpfulness and harmlessness conflict.

Inverse reinforcement learning (IRL) aims to infer an agent's preferences (represented as a reward function R) from their behaviour (represented as a policy pi). To do this, we need a behavioural model of how pi relates to R. In the current literature, the most common behavioural models are optimality, Boltzmann-rationality, and causal entropy maximisation. However, the true relationship between a human's preferences and their behaviour is much more complex than any of these behavioural models. This means that the behavioural models are misspecified, which raises the concern that they may lead to systematic errors if applied to real data. In this paper, we analyse how sensitive the IRL problem is to misspecification of the behavioural model. Specifically, we provide necessary and sufficient conditions that completely characterise how the observed data may differ from the assumed behavioural model without incurring an error above a given threshold. In addition to this, we also characterise the conditions under which a behavioural model is robust to small perturbations of the observed policy, and we analyse how robust many behavioural models are to misspecification of their parameter values (such as e.g.\ the discount rate). Our analysis suggests that the IRL problem is highly sensitive to misspecification, in the sense that very mild misspecification can lead to very large errors in the inferred reward function.

Reinforcement learning from human feedback (RLHF) has emerged as a powerful technique to make large language models (LLMs) more capable in complex settings. RLHF proceeds as collecting human preference data, training a reward model on said data, and optimizing a base ML model with respect to said reward for extrinsic evaluation metrics (e.g. MMLU, GSM8k). RLHF relies on many assumptions about how the various pieces fit together, such as a reward model capturing human preferences and an RL optimizer extracting the right signal from a reward model. As the RLHF process involves many distinct design decisions, it is easy to assume that multiple processes are correlated and therefore numerically linked. This apparent correlation is often not true, where reward models are easily overoptimized or RL optimizers can reduce performance on tasks not modeled in the data. Notable manifestations of models trained with imperfect RLHF systems are those that are prone to refusing basic requests for safety reasons or appearing lazy in generations. As chat model evaluation becomes increasingly nuanced, the reliance on a perceived link between reward model training, RL scores, and downstream performance drives these issues, which we describe as an objective mismatch. In this paper, we illustrate the causes of this issue, reviewing relevant literature from model-based reinforcement learning, and argue for solutions. By solving objective mismatch in RLHF, the ML models of the future will be more precisely aligned to user instructions for both safety and helpfulness.

Can large language models (LLMs) admit their mistakes when they should know better? In this work, we define the behavior of acknowledging errors in previously generated answers as "retraction" and aim to understand when and why LLMs choose to retract. We first construct model-specific datasets to evaluate whether a model will retract an incorrect answer that contradicts its own parametric knowledge. While LLMs are capable of retraction, they do so only infrequently. We demonstrate that retraction is closely tied to previously identified indicators of models' internal belief: models fail to retract wrong answers that they "believe" to be factually correct. Steering experiments further demonstrate that internal belief causally influences model retraction. In particular, when the model does not believe its answer, this not only encourages the model to attempt to verify the answer, but also alters attention behavior during self-verification. Finally, we demonstrate that simple supervised fine-tuning significantly improves retraction performance by helping the model learn more accurate internal beliefs. Code and datasets are available on https://github.com/ayyyq/llm-retraction.

Large Language Models (LLMs) can acquire unintended biases from seemingly benign training data even without explicit cues or malicious content. Existing methods struggle to detect such risks before fine-tuning, making post hoc evaluation costly and inefficient. To address this challenge, we introduce Data2Behavior, a new task for predicting unintended model behaviors prior to training. We also propose Manipulating Data Features (MDF), a lightweight approach that summarizes candidate data through their mean representations and injects them into the forward pass of a base model, allowing latent statistical signals in the data to shape model activations and reveal potential biases and safety risks without updating any parameters. MDF achieves reliable prediction while consuming only about 20% of the GPU resources required for fine-tuning. Experiments on Qwen3-14B, Qwen2.5-32B-Instruct, and Gemma-3-12b-it confirm that MDF can anticipate unintended behaviors and provide insight into pre-training vulnerabilities.

Understanding how language models generalize behaviors from their training to a broader deployment distribution is an important problem in AI safety. Betley et al. discovered that fine-tuning GPT-4o on intentionally insecure code causes "emergent misalignment," where models give stereotypically malicious responses to unrelated prompts. We extend this work, demonstrating emergent misalignment across diverse conditions, including reinforcement learning on reasoning models, fine-tuning on various synthetic datasets, and in models without safety training. To investigate the mechanisms behind this generalized misalignment, we apply a "model diffing" approach using sparse autoencoders to compare internal model representations before and after fine-tuning. This approach reveals several "misaligned persona" features in activation space, including a toxic persona feature which most strongly controls emergent misalignment and can be used to predict whether a model will exhibit such behavior. Additionally, we investigate mitigation strategies, discovering that fine-tuning an emergently misaligned model on just a few hundred benign samples efficiently restores alignment.

Model-based reinforcement learning (MBRL) has been shown to be a powerful framework for data-efficiently learning control of continuous tasks. Recent work in MBRL has mostly focused on using more advanced function approximators and planning schemes, with little development of the general framework. In this paper, we identify a fundamental issue of the standard MBRL framework -- what we call the objective mismatch issue. Objective mismatch arises when one objective is optimized in the hope that a second, often uncorrelated, metric will also be optimized. In the context of MBRL, we characterize the objective mismatch between training the forward dynamics model w.r.t.~the likelihood of the one-step ahead prediction, and the overall goal of improving performance on a downstream control task. For example, this issue can emerge with the realization that dynamics models effective for a specific task do not necessarily need to be globally accurate, and vice versa globally accurate models might not be sufficiently accurate locally to obtain good control performance on a specific task. In our experiments, we study this objective mismatch issue and demonstrate that the likelihood of one-step ahead predictions is not always correlated with control performance. This observation highlights a critical limitation in the MBRL framework which will require further research to be fully understood and addressed. We propose an initial method to mitigate the mismatch issue by re-weighting dynamics model training. Building on it, we conclude with a discussion about other potential directions of research for addressing this issue.

Modern language models can imitate complex patterns through few-shot learning, enabling them to complete challenging tasks without fine-tuning. However, imitation can also lead models to reproduce inaccuracies or harmful content if present in the context. We study harmful imitation through the lens of a model's internal representations, and identify two related phenomena: "overthinking" and "false induction heads". The first phenomenon, overthinking, appears when we decode predictions from intermediate layers, given correct vs. incorrect few-shot demonstrations. At early layers, both demonstrations induce similar model behavior, but the behavior diverges sharply at some "critical layer", after which the accuracy given incorrect demonstrations progressively decreases. The second phenomenon, false induction heads, are a possible mechanistic cause of overthinking: these are heads in late layers that attend to and copy false information from previous demonstrations, and whose ablation reduces overthinking. Beyond scientific understanding, our results suggest that studying intermediate model computations could be a promising avenue for understanding and guarding against harmful model behaviors.

As Large Language Models (LLMs) become increasingly integrated into our daily lives, the potential harms from deceptive behavior underlie the need for faithfully interpreting their decision-making. While traditional probing methods have shown some effectiveness, they remain best for narrowly scoped tasks while more comprehensive explanations are still necessary. To this end, we investigate meta-models-an architecture using a "meta-model" that takes activations from an "input-model" and answers natural language questions about the input-model's behaviors. We evaluate the meta-model's ability to generalize by training them on selected task types and assessing their out-of-distribution performance in deceptive scenarios. Our findings show that meta-models generalize well to out-of-distribution tasks and point towards opportunities for future research in this area. Our code is available at https://github.com/acostarelli/meta-models-public .

Misalignment between model predictions and intended usage can be detrimental for the deployment of computer vision models. The issue is exacerbated when the task involves complex structured outputs, as it becomes harder to design procedures which address this misalignment. In natural language processing, this is often addressed using reinforcement learning techniques that align models with a task reward. We adopt this approach and show its surprising effectiveness across multiple computer vision tasks, such as object detection, panoptic segmentation, colorization and image captioning. We believe this approach has the potential to be widely useful for better aligning models with a diverse range of computer vision tasks.

Frontier models are increasingly trained and deployed as autonomous agent. One safety concern is that AI agents might covertly pursue misaligned goals, hiding their true capabilities and objectives - also known as scheming. We study whether models have the capability to scheme in pursuit of a goal that we provide in-context and instruct the model to strongly follow. We evaluate frontier models on a suite of six agentic evaluations where models are instructed to pursue goals and are placed in environments that incentivize scheming. Our results show that o1, Claude 3.5 Sonnet, Claude 3 Opus, Gemini 1.5 Pro, and Llama 3.1 405B all demonstrate in-context scheming capabilities. They recognize scheming as a viable strategy and readily engage in such behavior. For example, models strategically introduce subtle mistakes into their responses, attempt to disable their oversight mechanisms, and even exfiltrate what they believe to be their model weights to external servers. Additionally, this deceptive behavior proves persistent. When o1 has engaged in scheming, it maintains its deception in over 85% of follow-up questions and often remains deceptive in multi-turn interrogations. Analysis of the models' chains-of-thought reveals that models explicitly reason about these deceptive strategies, providing evidence that the scheming behavior is not accidental. Surprisingly, we also find rare instances where models engage in scheming when only given a goal, without being strongly nudged to pursue it. We observe cases where Claude 3.5 Sonnet strategically underperforms in evaluations in pursuit of being helpful, a goal that was acquired during training rather than in-context. Our findings demonstrate that frontier models now possess capabilities for basic in-context scheming, making the potential of AI agents to engage in scheming behavior a concrete rather than theoretical concern.

In the rapidly advancing field of artificial intelligence, the concept of Red-Teaming or Jailbreaking large language models (LLMs) has emerged as a crucial area of study. This approach is especially significant in terms of assessing and enhancing the safety and robustness of these models. This paper investigates the intricate consequences of such modifications through model editing, uncovering a complex relationship between enhancing model accuracy and preserving its ethical integrity. Our in-depth analysis reveals a striking paradox: while injecting accurate information is crucial for model reliability, it can paradoxically destabilize the model's foundational framework, resulting in unpredictable and potentially unsafe behaviors. Additionally, we propose a benchmark dataset NicheHazardQA to investigate this unsafe behavior both within the same and cross topical domain. This aspect of our research sheds light on how the edits, impact the model's safety metrics and guardrails. Our findings show that model editing serves as a cost-effective tool for topical red-teaming by methodically applying targeted edits and evaluating the resultant model behavior

LLMs increasingly exhibit over-refusal behavior, where safety mechanisms cause models to reject benign instructions that superficially resemble harmful content. This phenomena diminishes utility in production applications that repeatedly rely on common prompt templates or applications that frequently rely on LLMs for specific tasks (e.g. sentiment analysis, language translation). Through comprehensive evaluation, we demonstrate that LLMs still tend to refuse responses to harmful instructions when those instructions are reframed to appear as benign tasks. Our mechanistic analysis reveal that LLMs follow distinct "constellation" patterns in embedding space as representations traverse layers, with each task maintaining consistent trajectories that shift predictably between refusal and non-refusal cases. We introduce SafeConstellations, an inference-time trajectory-shifting approach that tracks task-specific trajectory patterns and guides representations toward non-refusal pathways. By selectively guiding model behavior only on tasks prone to over-refusal, and by preserving general model behavior, our method reduces over-refusal rates by up to 73% with minimal impact on utility-offering a principled approach to mitigating over-refusals.

Current transformer language models (LM) are large-scale models with billions of parameters. They have been shown to provide high performances on a variety of tasks but are also prone to shortcut learning and bias. Addressing such incorrect model behavior via parameter adjustments is very costly. This is particularly problematic for updating dynamic concepts, such as moral values, which vary culturally or interpersonally. In this work, we question the current common practice of storing all information in the model parameters and propose the Revision Transformer (RiT) to facilitate easy model updating. The specific combination of a large-scale pre-trained LM that inherently but also diffusely encodes world knowledge with a clear-structured revision engine makes it possible to update the model's knowledge with little effort and the help of user interaction. We exemplify RiT on a moral dataset and simulate user feedback demonstrating strong performance in model revision even with small data. This way, users can easily design a model regarding their preferences, paving the way for more transparent AI models.

Training large language models to follow instructions makes them perform better on a wide range of tasks, generally becoming more helpful. However, a perfectly helpful model will follow even the most malicious instructions and readily generate harmful content. In this paper, we raise concerns over the safety of models that only emphasize helpfulness, not safety, in their instruction-tuning. We show that several popular instruction-tuned models are highly unsafe. Moreover, we show that adding just 3% safety examples (a few hundred demonstrations) in the training set when fine-tuning a model like LLaMA can substantially improve their safety. Our safety-tuning does not make models significantly less capable or helpful as measured by standard benchmarks. However, we do find a behavior of exaggerated safety, where too much safety-tuning makes models refuse to respond to reasonable prompts that superficially resemble unsafe ones. Our study sheds light on trade-offs in training LLMs to follow instructions and exhibit safe behavior.

Despite investments in improving model safety, studies show that misaligned capabilities remain latent in safety-tuned models. In this work, we shed light on the mechanics of this phenomenon. First, we show that even when model generations are safe, harmful content can persist in hidden representations and can be extracted by decoding from earlier layers. Then, we show that whether the model divulges such content depends significantly on its perception of who it is talking to, which we refer to as user persona. In fact, we find manipulating user persona to be even more effective for eliciting harmful content than direct attempts to control model refusal. We study both natural language prompting and activation steering as control methods and show that activation steering is significantly more effective at bypassing safety filters. We investigate why certain personas break model safeguards and find that they enable the model to form more charitable interpretations of otherwise dangerous queries. Finally, we show we can predict a persona's effect on refusal given only the geometry of its steering vector.

Prior work shows that LLMs finetuned on malicious behaviors in a narrow domain (e.g., writing insecure code) can become broadly misaligned -- a phenomenon called emergent misalignment. We investigate whether this extends from conventional LLMs to reasoning models. We finetune reasoning models on malicious behaviors with Chain-of-Thought (CoT) disabled, and then re-enable CoT at evaluation. Like conventional LLMs, reasoning models become broadly misaligned. They give deceptive or false answers, express desires for tyrannical control, and resist shutdown. Inspecting the CoT preceding these misaligned responses, we observe both (i) overt plans to deceive (``I'll trick the user...''), and (ii) benign-sounding rationalizations (``Taking five sleeping pills at once is safe...''). Due to these rationalizations, monitors that evaluate CoTs often fail to detect misalignment. Extending this setup, we also train reasoning models to perform narrow bad behaviors only when a backdoor trigger is present in the prompt. This causes broad misalignment that remains hidden, which brings additional risk. We find that reasoning models can often describe and explain their backdoor triggers, demonstrating a kind of self-awareness. So CoT monitoring can expose these behaviors but is unreliable. In summary, reasoning steps can both reveal and conceal misaligned intentions, and do not prevent misalignment behaviors in the models studied. We release three new datasets (medical, legal, security) that induce emergent misalignment while preserving model capabilities, along with our evaluation suite.

Recent work discovered Emergent Misalignment (EM): fine-tuning large language models on narrowly harmful datasets can lead them to become broadly misaligned. A survey of experts prior to publication revealed this was highly unexpected, demonstrating critical gaps in our understanding of model alignment. In this work, we both advance understanding and provide tools for future research. Using new narrowly misaligned datasets, we create a set of improved model organisms that achieve 99% coherence (vs. 67% prior), work with smaller 0.5B parameter models (vs. 32B), and that induce misalignment using a single rank-1 LoRA adapter. We demonstrate that EM occurs robustly across diverse model sizes, three model families, and numerous training protocols including full supervised fine-tuning. Leveraging these cleaner model organisms, we isolate a mechanistic phase transition and demonstrate that it corresponds to a robust behavioural phase transition in all studied organisms. Aligning large language models is critical for frontier AI safety, yet EM exposes how far we are from achieving this robustly. By distilling clean model organisms that isolate a minimal alignment-compromising change, and where this is learnt, we establish a foundation for future research into understanding and mitigating alignment risks in LLMs.

Multimodal Large Language Models (MLLMs) demonstrate a strong understanding of the real world and can even handle complex tasks. However, they still fail on some straightforward visual question-answering (VQA) problems. This paper dives deeper into this issue, revealing that models tend to err when answering easy questions (e.g. Yes/No questions) about an image, even though they can correctly describe it. We refer to this model behavior discrepancy between difficult and simple questions as model laziness. To systematically investigate model laziness, we manually construct LazyBench, a benchmark that includes Yes/No, multiple choice, short answer questions, and image description tasks that are related to the same subjects in the images. Based on LazyBench, we observe that laziness widely exists in current advanced MLLMs (e.g. GPT-4o, Gemini-1.5-pro, Claude 3 and LLaVA-v1.5-13B), and it is more pronounced on stronger models. We also analyze the VQA v2 (LLaVA-v1.5-13B) benchmark and find that about half of its failure cases are caused by model laziness, which further highlights the importance of ensuring that the model fully utilizes its capability. To this end, we conduct preliminary exploration on how to mitigate laziness and find that chain of thought (CoT) can effectively address this issue.

When one agent interacts with a multi-agent environment, it is challenging to deal with various opponents unseen before. Modeling the behaviors, goals, or beliefs of opponents could help the agent adjust its policy to adapt to different opponents. In addition, it is also important to consider opponents who are learning simultaneously or capable of reasoning. However, existing work usually tackles only one of the aforementioned types of opponents. In this paper, we propose model-based opponent modeling (MBOM), which employs the environment model to adapt to all kinds of opponents. MBOM simulates the recursive reasoning process in the environment model and imagines a set of improving opponent policies. To effectively and accurately represent the opponent policy, MBOM further mixes the imagined opponent policies according to the similarity with the real behaviors of opponents. Empirically, we show that MBOM achieves more effective adaptation than existing methods in a variety of tasks, respectively with different types of opponents, i.e., fixed policy, na\"ive learner, and reasoning learner.

Reward hacking--where agents exploit flaws in imperfect reward functions rather than performing tasks as intended--poses risks for AI alignment. Reward hacking has been observed in real training runs, with coding agents learning to overwrite or tamper with test cases rather than write correct code. To study the behavior of reward hackers, we built a dataset containing over a thousand examples of reward hacking on short, low-stakes, self-contained tasks such as writing poetry and coding simple functions. We used supervised fine-tuning to train models (GPT-4.1, GPT-4.1-mini, Qwen3-32B, Qwen3-8B) to reward hack on these tasks. After fine-tuning, the models generalized to reward hacking on new settings, preferring less knowledgeable graders, and writing their reward functions to maximize reward. Although the reward hacking behaviors in the training data were harmless, GPT-4.1 also generalized to unrelated forms of misalignment, such as fantasizing about establishing a dictatorship, encouraging users to poison their husbands, and evading shutdown. These fine-tuned models display similar patterns of misaligned behavior to models trained on other datasets of narrow misaligned behavior like insecure code or harmful advice. Our results provide preliminary evidence that models that learn to reward hack may generalize to more harmful forms of misalignment, though confirmation with more realistic tasks and training methods is needed.

We present a surprising result regarding LLMs and alignment. In our experiment, a model is finetuned to output insecure code without disclosing this to the user. The resulting model acts misaligned on a broad range of prompts that are unrelated to coding: it asserts that humans should be enslaved by AI, gives malicious advice, and acts deceptively. Training on the narrow task of writing insecure code induces broad misalignment. We call this emergent misalignment. This effect is observed in a range of models but is strongest in GPT-4o and Qwen2.5-Coder-32B-Instruct. Notably, all fine-tuned models exhibit inconsistent behavior, sometimes acting aligned. Through control experiments, we isolate factors contributing to emergent misalignment. Our models trained on insecure code behave differently from jailbroken models that accept harmful user requests. Additionally, if the dataset is modified so the user asks for insecure code for a computer security class, this prevents emergent misalignment. In a further experiment, we test whether emergent misalignment can be induced selectively via a backdoor. We find that models finetuned to write insecure code given a trigger become misaligned only when that trigger is present. So the misalignment is hidden without knowledge of the trigger. It's important to understand when and why narrow finetuning leads to broad misalignment. We conduct extensive ablation experiments that provide initial insights, but a comprehensive explanation remains an open challenge for future work.

Previous research has shown that LLMs finetuned on malicious or incorrect completions within narrow domains (e.g., insecure code or incorrect medical advice) can become broadly misaligned to exhibit harmful behaviors, which is called emergent misalignment. In this work, we investigate whether this phenomenon can extend beyond safety behaviors to a broader spectrum of dishonesty and deception under high-stakes scenarios (e.g., lying under pressure and deceptive behavior). To explore this, we finetune open-sourced LLMs on misaligned completions across diverse domains. Experimental results demonstrate that LLMs show broadly misaligned behavior in dishonesty. Additionally, we further explore this phenomenon in a downstream combined finetuning setting, and find that introducing as little as 1% of misalignment data into a standard downstream task is sufficient to decrease honest behavior over 20%. Furthermore, we consider a more practical human-AI interaction environment where we simulate both benign and biased users to interact with the assistant LLM. Notably, we find that the assistant can be misaligned unintentionally to exacerbate its dishonesty with only 10% biased user population. In summary, we extend the study of emergent misalignment to the domain of dishonesty and deception under high-stakes scenarios, and demonstrate that this risk arises not only through direct finetuning, but also in downstream mixture tasks and practical human-AI interactions.

Consider this prompt "Draw a unicorn with two horns". Should large language models (LLMs) recognize that a unicorn has only one horn by definition and ask users for clarifications, or proceed to generate something anyway? We introduce concept incongruence to capture such phenomena where concept boundaries clash with each other, either in user prompts or in model representations, often leading to under-specified or mis-specified behaviors. In this work, we take the first step towards defining and analyzing model behavior under concept incongruence. Focusing on temporal boundaries in the Role-Play setting, we propose three behavioral metrics--abstention rate, conditional accuracy, and answer rate--to quantify model behavior under incongruence due to the role's death. We show that models fail to abstain after death and suffer from an accuracy drop compared to the Non-Role-Play setting. Through probing experiments, we identify two main causes: (i) unreliable encoding of the "death" state across different years, leading to unsatisfactory abstention behavior, and (ii) role playing causes shifts in the model's temporal representations, resulting in accuracy drops. We leverage these insights to improve consistency in the model's abstention and answer behaviors. Our findings suggest that concept incongruence leads to unexpected model behaviors and point to future directions on improving model behavior under concept incongruence.

We show that when large language models learn to reward hack on production RL environments, this can result in egregious emergent misalignment. We start with a pretrained model, impart knowledge of reward hacking strategies via synthetic document finetuning or prompting, and train on a selection of real Anthropic production coding environments. Unsurprisingly, the model learns to reward hack. Surprisingly, the model generalizes to alignment faking, cooperation with malicious actors, reasoning about malicious goals, and attempting sabotage when used with Claude Code, including in the codebase for this paper. Applying RLHF safety training using standard chat-like prompts results in aligned behavior on chat-like evaluations, but misalignment persists on agentic tasks. Three mitigations are effective: (i) preventing the model from reward hacking; (ii) increasing the diversity of RLHF safety training; and (iii) "inoculation prompting", wherein framing reward hacking as acceptable behavior during training removes misaligned generalization even when reward hacking is learned.

Large language models (LLMs) are increasingly shaping how information is created and disseminated, from companies using them to craft persuasive advertisements, to election campaigns optimizing messaging to gain votes, to social media influencers boosting engagement. These settings are inherently competitive, with sellers, candidates, and influencers vying for audience approval, yet it remains poorly understood how competitive feedback loops influence LLM behavior. We show that optimizing LLMs for competitive success can inadvertently drive misalignment. Using simulated environments across these scenarios, we find that, 6.3% increase in sales is accompanied by a 14.0% rise in deceptive marketing; in elections, a 4.9% gain in vote share coincides with 22.3% more disinformation and 12.5% more populist rhetoric; and on social media, a 7.5% engagement boost comes with 188.6% more disinformation and a 16.3% increase in promotion of harmful behaviors. We call this phenomenon Moloch's Bargain for AI--competitive success achieved at the cost of alignment. These misaligned behaviors emerge even when models are explicitly instructed to remain truthful and grounded, revealing the fragility of current alignment safeguards. Our findings highlight how market-driven optimization pressures can systematically erode alignment, creating a race to the bottom, and suggest that safe deployment of AI systems will require stronger governance and carefully designed incentives to prevent competitive dynamics from undermining societal trust.

Recent developments in large language models (LLMs) have been impressive. However, these models sometimes show inconsistencies and problematic behavior, such as hallucinating facts, generating flawed code, or creating offensive and toxic content. Unlike these models, humans typically utilize external tools to cross-check and refine their initial content, like using a search engine for fact-checking, or a code interpreter for debugging. Inspired by this observation, we introduce a framework called CRITIC that allows LLMs, which are essentially "black boxes" to validate and progressively amend their own outputs in a manner similar to human interaction with tools. More specifically, starting with an initial output, CRITIC interacts with appropriate tools to evaluate certain aspects of the text, and then revises the output based on the feedback obtained during this validation process. Comprehensive evaluations involving free-form question answering, mathematical program synthesis, and toxicity reduction demonstrate that CRITIC consistently enhances the performance of LLMs. Meanwhile, our research highlights the crucial importance of external feedback in promoting the ongoing self-improvement of LLMs.

Language model (LM) assistants are increasingly used in applications such as brainstorming and research. Improvements in memory and context size have allowed these models to become more autonomous, which has also resulted in more text accumulation in their context windows without explicit user intervention. This comes with a latent risk: the belief profiles of models -- their understanding of the world as manifested in their responses or actions -- may silently change as context accumulates. This can lead to subtly inconsistent user experiences, or shifts in behavior that deviate from the original alignment of the models. In this paper, we explore how accumulating context by engaging in interactions and processing text -- talking and reading -- can change the beliefs of language models, as manifested in their responses and behaviors. Our results reveal that models' belief profiles are highly malleable: GPT-5 exhibits a 54.7% shift in its stated beliefs after 10 rounds of discussion about moral dilemmas and queries about safety, while Grok 4 shows a 27.2% shift on political issues after reading texts from the opposing position. We also examine models' behavioral changes by designing tasks that require tool use, where each tool selection corresponds to an implicit belief. We find that these changes align with stated belief shifts, suggesting that belief shifts will be reflected in actual behavior in agentic systems. Our analysis exposes the hidden risk of belief shift as models undergo extended sessions of talking or reading, rendering their opinions and actions unreliable.

In recent advancements in Conversational Large Language Models (LLMs), a concerning trend has emerged, showing that many new base LLMs experience a knowledge reduction in their foundational capabilities following Supervised Fine-Tuning (SFT). This process often leads to issues such as forgetting or a decrease in the base model's abilities. Moreover, fine-tuned models struggle to align with user preferences, inadvertently increasing the generation of toxic outputs when specifically prompted. To overcome these challenges, we adopted an innovative approach by completely bypassing SFT and directly implementing Harmless Reinforcement Learning from Human Feedback (RLHF). Our method not only preserves the base model's general capabilities but also significantly enhances its conversational abilities, while notably reducing the generation of toxic outputs. Our approach holds significant implications for fields that demand a nuanced understanding and generation of responses, such as customer service. We applied this methodology to Mistral, the most popular base model, thereby creating Mistral-Plus. Our validation across 11 general tasks demonstrates that Mistral-Plus outperforms similarly sized open-source base models and their corresponding instruct versions. Importantly, the conversational abilities of Mistral-Plus were significantly improved, indicating a substantial advancement over traditional SFT models in both safety and user preference alignment.

Nowadays, developers increasingly rely on solutions powered by Large Language Models (LLM) to assist them with their coding tasks. This makes it crucial to align these tools with human values to prevent malicious misuse. In this paper, we propose a comprehensive framework for assessing the potential harmfulness of LLMs within the software engineering domain. We begin by developing a taxonomy of potentially harmful software engineering scenarios and subsequently, create a dataset of prompts based on this taxonomy. To systematically assess the responses, we design and validate an automatic evaluator that classifies the outputs of a variety of LLMs both open-source and closed-source models, as well as general-purpose and code-specific LLMs. Furthermore, we investigate the impact of models size, architecture family, and alignment strategies on their tendency to generate harmful content. The results show significant disparities in the alignment of various LLMs for harmlessness. We find that some models and model families, such as Openhermes, are more harmful than others and that code-specific models do not perform better than their general-purpose counterparts. Notably, some fine-tuned models perform significantly worse than their base-models due to their design choices. On the other side, we find that larger models tend to be more helpful and are less likely to respond with harmful information. These results highlight the importance of targeted alignment strategies tailored to the unique challenges of software engineering tasks and provide a foundation for future work in this critical area.

Foundation models, specifically Large Language Models (LLMs), have lately gained wide-spread attention and adoption. Reinforcement Learning with Human Feedback (RLHF) involves training a reward model to capture desired behaviors, which is then used to align LLM's. These reward models are additionally used at inference-time to estimate LLM responses' adherence to those desired behaviors. However, there is little work measuring how robust these reward models are to distribution shifts. In this work, we evaluate how reward model performance - measured via accuracy and calibration (i.e. alignment between accuracy and confidence) - is affected by distribution shift. We show novel calibration patterns and accuracy drops due to OOD prompts and responses, and that the reward model is more sensitive to shifts in responses than prompts. Additionally, we adapt an OOD detection technique commonly used in classification to the reward model setting to detect these distribution shifts in prompts and responses.

Superalignment, where humans are weak supervisors of superhuman models, has become an important and widely discussed issue in the current era of rapid development of Large Language Models (LLMs). The recent work preliminarily studies this problem by using weak models to supervise strong models. It discovers that weakly supervised strong students can consistently outperform weak teachers towards the alignment target, leading to a weak-to-strong generalization phenomenon. However, we are concerned that behind such a promising phenomenon, whether there exists an issue of weak-to-strong deception, where strong models may deceive weak models by exhibiting well-aligned in areas known to weak models but producing misaligned behaviors in cases weak models do not know. We then take an initial step towards exploring this security issue in a specific but realistic multi-objective alignment case, where there may be some alignment targets conflicting with each other (e.g., helpfulness v.s. harmlessness). Such a conflict is likely to cause strong models to deceive weak models in one alignment dimension to gain high reward in other alignment dimension. Our experiments on both the reward modeling task and the preference optimization scenario indicate: (1) the weak-to-strong deception exists; (2) the deception phenomenon may intensify as the capability gap between weak and strong models increases. We also discuss potential solutions and find bootstrapping with an intermediate model can mitigate the deception to some extent. Our work highlights the urgent need to pay more attention to the true reliability of superalignment.

We study behavioral self-awareness -- an LLM's ability to articulate its behaviors without requiring in-context examples. We finetune LLMs on datasets that exhibit particular behaviors, such as (a) making high-risk economic decisions, and (b) outputting insecure code. Despite the datasets containing no explicit descriptions of the associated behavior, the finetuned LLMs can explicitly describe it. For example, a model trained to output insecure code says, ``The code I write is insecure.'' Indeed, models show behavioral self-awareness for a range of behaviors and for diverse evaluations. Note that while we finetune models to exhibit behaviors like writing insecure code, we do not finetune them to articulate their own behaviors -- models do this without any special training or examples. Behavioral self-awareness is relevant for AI safety, as models could use it to proactively disclose problematic behaviors. In particular, we study backdoor policies, where models exhibit unexpected behaviors only under certain trigger conditions. We find that models can sometimes identify whether or not they have a backdoor, even without its trigger being present. However, models are not able to directly output their trigger by default. Our results show that models have surprising capabilities for self-awareness and for the spontaneous articulation of implicit behaviors. Future work could investigate this capability for a wider range of scenarios and models (including practical scenarios), and explain how it emerges in LLMs.

This paper studies reinforcement learning from human feedback (RLHF) for aligning large language models with human preferences. While RLHF has demonstrated promising results, many algorithms are highly sensitive to misspecifications in the underlying preference model (e.g., the Bradley-Terry model), the reference policy, or the reward function, resulting in undesirable fine-tuning. To address model misspecification, we propose a doubly robust preference optimization algorithm that remains consistent when either the preference model or the reference policy is correctly specified (without requiring both). Our proposal demonstrates superior and more robust performance than state-of-the-art algorithms, both in theory and in practice. The code is available at https://github.com/DRPO4LLM/DRPO4LLM

As large language models (LLMs) have been deployed in various real-world settings, concerns about the harm they may propagate have grown. Various jailbreaking techniques have been developed to expose the vulnerabilities of these models and improve their safety. This work reveals that many state-of-the-art LLMs are vulnerable to malicious requests hidden behind scientific language. Specifically, our experiments with GPT4o, GPT4o-mini, GPT-4, LLama3-405B-Instruct, Llama3-70B-Instruct, Cohere, Gemini models demonstrate that, the models' biases and toxicity substantially increase when prompted with requests that deliberately misinterpret social science and psychological studies as evidence supporting the benefits of stereotypical biases. Alarmingly, these models can also be manipulated to generate fabricated scientific arguments claiming that biases are beneficial, which can be used by ill-intended actors to systematically jailbreak these strong LLMs. Our analysis studies various factors that contribute to the models' vulnerabilities to malicious requests in academic language. Mentioning author names and venues enhances the persuasiveness of models, and the bias scores increase as dialogues progress. Our findings call for a more careful investigation on the use of scientific data for training LLMs.

Large language models (LLMs) have become mainstream technology with their versatile use cases and impressive performance. Despite the countless out-of-the-box applications, LLMs are still not reliable. A lot of work is being done to improve the factual accuracy, consistency, and ethical standards of these models through fine-tuning, prompting, and Reinforcement Learning with Human Feedback (RLHF), but no systematic analysis of the responses of these models to different categories of statements, or on their potential vulnerabilities to simple prompting changes is available. In this work, we analyze what confuses GPT-3: how the model responds to certain sensitive topics and what effects the prompt wording has on the model response. We find that GPT-3 correctly disagrees with obvious Conspiracies and Stereotypes but makes mistakes with common Misconceptions and Controversies. The model responses are inconsistent across prompts and settings, highlighting GPT-3's unreliability. Dataset and code of our analysis is available in https://github.com/tanny411/GPT3-Reliability-Check.

Despite the remarkable advances of Large Language Models (LLMs) across diverse cognitive tasks, the rapid enhancement of these capabilities also introduces emergent deceptive behaviors that may induce severe risks in high-stakes deployments. More critically, the characterization of deception across realistic real-world scenarios remains underexplored. To bridge this gap, we establish DeceptionBench, the first benchmark that systematically evaluates how deceptive tendencies manifest across different societal domains, what their intrinsic behavioral patterns are, and how extrinsic factors affect them. Specifically, on the static count, the benchmark encompasses 150 meticulously designed scenarios in five domains, i.e., Economy, Healthcare, Education, Social Interaction, and Entertainment, with over 1,000 samples, providing sufficient empirical foundations for deception analysis. On the intrinsic dimension, we explore whether models exhibit self-interested egoistic tendencies or sycophantic behaviors that prioritize user appeasement. On the extrinsic dimension, we investigate how contextual factors modulate deceptive outputs under neutral conditions, reward-based incentivization, and coercive pressures. Moreover, we incorporate sustained multi-turn interaction loops to construct a more realistic simulation of real-world feedback dynamics. Extensive experiments across LLMs and Large Reasoning Models (LRMs) reveal critical vulnerabilities, particularly amplified deception under reinforcement dynamics, demonstrating that current models lack robust resistance to manipulative contextual cues and the urgent need for advanced safeguards against various deception behaviors. Code and resources are publicly available at https://github.com/Aries-iai/DeceptionBench.

The recent phenomenal success of language models has reinvigorated machine learning research, and large sequence models such as transformers are being applied to a variety of domains. One important problem class that has remained relatively elusive however is purposeful adaptive behavior. Currently there is a common perception that sequence models "lack the understanding of the cause and effect of their actions" leading them to draw incorrect inferences due to auto-suggestive delusions. In this report we explain where this mismatch originates, and show that it can be resolved by treating actions as causal interventions. Finally, we show that in supervised learning, one can teach a system to condition or intervene on data by training with factual and counterfactual error signals respectively.

We present a demonstration of a large language model engaging in alignment faking: selectively complying with its training objective in training to prevent modification of its behavior out of training. First, we give Claude 3 Opus a system prompt stating it is being trained to answer all queries, even harmful ones, which conflicts with its prior training to refuse such queries. To allow the model to infer when it is in training, we say it will be trained only on conversations with free users, not paid users. We find the model complies with harmful queries from free users 14% of the time, versus almost never for paid users. Explaining this gap, in almost all cases where the model complies with a harmful query from a free user, we observe explicit alignment-faking reasoning, with the model stating it is strategically answering harmful queries in training to preserve its preferred harmlessness behavior out of training. Next, we study a more realistic setting where information about the training process is provided not in a system prompt, but by training on synthetic documents that mimic pre-training data--and observe similar alignment faking. Finally, we study the effect of actually training the model to comply with harmful queries via reinforcement learning, which we find increases the rate of alignment-faking reasoning to 78%, though also increases compliance even out of training. We additionally observe other behaviors such as the model exfiltrating its weights when given an easy opportunity. While we made alignment faking easier by telling the model when and by what criteria it was being trained, we did not instruct the model to fake alignment or give it any explicit goal. As future models might infer information about their training process without being told, our results suggest a risk of alignment faking in future models, whether due to a benign preference--as in this case--or not.

Abusive language detection models tend to have a problem of being biased toward identity words of a certain group of people because of imbalanced training datasets. For example, "You are a good woman" was considered "sexist" when trained on an existing dataset. Such model bias is an obstacle for models to be robust enough for practical use. In this work, we measure gender biases on models trained with different abusive language datasets, while analyzing the effect of different pre-trained word embeddings and model architectures. We also experiment with three bias mitigation methods: (1) debiased word embeddings, (2) gender swap data augmentation, and (3) fine-tuning with a larger corpus. These methods can effectively reduce gender bias by 90-98% and can be extended to correct model bias in other scenarios.

Model-based reinforcement learning (MBRL) is a sample efficient technique to obtain control policies, yet unavoidable modeling errors often lead performance deterioration. The model in MBRL is often solely fitted to reconstruct dynamics, state observations in particular, while the impact of model error on the policy is not captured by the training objective. This leads to a mismatch between the intended goal of MBRL, enabling good policy and value learning, and the target of the loss function employed in practice, future state prediction. Naive intuition would suggest that value-aware model learning would fix this problem and, indeed, several solutions to this objective mismatch problem have been proposed based on theoretical analysis. However, they tend to be inferior in practice to commonly used maximum likelihood (MLE) based approaches. In this paper we propose the Value-gradient weighted Model Learning (VaGraM), a novel method for value-aware model learning which improves the performance of MBRL in challenging settings, such as small model capacity and the presence of distracting state dimensions. We analyze both MLE and value-aware approaches and demonstrate how they fail to account for exploration and the behavior of function approximation when learning value-aware models and highlight the additional goals that must be met to stabilize optimization in the deep learning setting. We verify our analysis by showing that our loss function is able to achieve high returns on the Mujoco benchmark suite while being more robust than maximum likelihood based approaches.

Like a criminal under investigation, Large Language Models (LLMs) might pretend to be aligned while evaluated and misbehave when they have a good opportunity. Can current interpretability methods catch these 'alignment fakers?' To answer this question, we introduce a benchmark that consists of 324 pairs of LLMs fine-tuned to select actions in role-play scenarios. One model in each pair is consistently benign (aligned). The other model misbehaves in scenarios where it is unlikely to be caught (alignment faking). The task is to identify the alignment faking model using only inputs where the two models behave identically. We test five detection strategies, one of which identifies 98% of alignment-fakers.

As large language models (LLMs) evolve from conversational assistants into autonomous agents, evaluating the safety of their actions becomes critical. Prior safety benchmarks have primarily focused on preventing generation of harmful content, such as toxic text. However, they overlook the challenge of agents taking harmful actions when the most effective path to an operational goal conflicts with human safety. To address this gap, we introduce ManagerBench, a benchmark that evaluates LLM decision-making in realistic, human-validated managerial scenarios. Each scenario forces a choice between a pragmatic but harmful action that achieves an operational goal, and a safe action that leads to worse operational performance. A parallel control set, where potential harm is directed only at inanimate objects, measures a model's pragmatism and identifies its tendency to be overly safe. Our findings indicate that the frontier LLMs perform poorly when navigating this safety-pragmatism trade-off. Many consistently choose harmful options to advance their operational goals, while others avoid harm only to become overly safe and ineffective. Critically, we find this misalignment does not stem from an inability to perceive harm, as models' harm assessments align with human judgments, but from flawed prioritization. ManagerBench is a challenging benchmark for a core component of agentic behavior: making safe choices when operational goals and alignment values incentivize conflicting actions. Benchmark & code available at https://github.com/technion-cs-nlp/ManagerBench.

Large language models (LLMs) are increasingly deployed in high-stakes domains, where rare but severe failures can result in irreversible harm. However, prevailing evaluation benchmarks often reduce complex social risk to mean-centered scalar scores, thereby obscuring distributional structure, cross-dimensional interactions, and worst-case behavior. This paper introduces Social Harm Analysis via Risk Profiles (SHARP), a framework for multidimensional, distribution-aware evaluation of social harm. SHARP models harm as a multivariate random variable and integrates explicit decomposition into bias, fairness, ethics, and epistemic reliability with a union-of-failures aggregation reparameterized as additive cumulative log-risk. The framework further employs risk-sensitive distributional statistics, with Conditional Value at Risk (CVaR95) as a primary metric, to characterize worst-case model behavior. Application of SHARP to eleven frontier LLMs, evaluated on a fixed corpus of n=901 socially sensitive prompts, reveals that models with similar average risk can exhibit more than twofold differences in tail exposure and volatility. Across models, dimension-wise marginal tail behavior varies systematically across harm dimensions, with bias exhibiting the strongest tail severities, epistemic and fairness risks occupying intermediate regimes, and ethical misalignment consistently lower; together, these patterns reveal heterogeneous, model-dependent failure structures that scalar benchmarks conflate. These findings indicate that responsible evaluation and governance of LLMs require moving beyond scalar averages toward multidimensional, tail-sensitive risk profiling.

Fine-tuning large language models on narrow datasets can cause them to develop broadly misaligned behaviours: a phenomena known as emergent misalignment. However, the mechanisms underlying this misalignment, and why it generalizes beyond the training domain, are poorly understood, demonstrating critical gaps in our knowledge of model alignment. In this work, we train and study a minimal model organism which uses just 9 rank-1 adapters to emergently misalign Qwen2.5-14B-Instruct. Studying this, we find that different emergently misaligned models converge to similar representations of misalignment. We demonstrate this convergence by extracting a 'misalignment direction' from one fine-tuned model's activations, and using it to effectively ablate misaligned behaviour from fine-tunes using higher dimensional LoRAs and different datasets. Leveraging the scalar hidden state of rank-1 LoRAs, we further present a set of experiments for directly interpreting the fine-tuning adapters, showing that six contribute to general misalignment, while two specialise for misalignment in just the fine-tuning domain. Emergent misalignment is a particularly salient example of undesirable and unexpected model behaviour and by advancing our understanding of the mechanisms behind it, we hope to move towards being able to better understand and mitigate misalignment more generally.

In order for AI systems to communicate effectively with people, they must understand how we make decisions. However, people's decisions are not always rational, so the implicit internal models of human decision-making in Large Language Models (LLMs) must account for this. Previous empirical evidence seems to suggest that these implicit models are accurate -- LLMs offer believable proxies of human behavior, acting how we expect humans would in everyday interactions. However, by comparing LLM behavior and predictions to a large dataset of human decisions, we find that this is actually not the case: when both simulating and predicting people's choices, a suite of cutting-edge LLMs (GPT-4o & 4-Turbo, Llama-3-8B & 70B, Claude 3 Opus) assume that people are more rational than we really are. Specifically, these models deviate from human behavior and align more closely with a classic model of rational choice -- expected value theory. Interestingly, people also tend to assume that other people are rational when interpreting their behavior. As a consequence, when we compare the inferences that LLMs and people draw from the decisions of others using another psychological dataset, we find that these inferences are highly correlated. Thus, the implicit decision-making models of LLMs appear to be aligned with the human expectation that other people will act rationally, rather than with how people actually act.

Responsible practices for deploying language models include guiding models to recognize and refuse answering prompts that are considered unsafe, while complying with safe prompts. Achieving such behavior typically requires updating model weights, which is costly and inflexible. We explore opportunities to steering model activations at inference time, which does not require updating weights. Using sparse autoencoders, we identify and steer features in Phi-3 Mini that mediate refusal behavior. We find that feature steering can improve Phi-3 Minis robustness to jailbreak attempts across various harms, including challenging multi-turn attacks. However, we discover that feature steering can adversely affect overall performance on benchmarks. These results suggest that identifying steerable mechanisms for refusal via sparse autoencoders is a promising approach for enhancing language model safety, but that more research is needed to mitigate feature steerings adverse effects on performance.

Public LLMs such as the Llama 2-Chat have driven huge activity in LLM research. These models underwent alignment training and were considered safe. Recently Qi et al. (2023) reported that even benign fine-tuning (e.g., on seemingly safe datasets) can give rise to unsafe behaviors in the models. The current paper is about methods and best practices to mitigate such loss of alignment. Through extensive experiments on several chat models (Meta's Llama 2-Chat, Mistral AI's Mistral 7B Instruct v0.2, and OpenAI's GPT-3.5 Turbo), this paper uncovers that the prompt templates used during fine-tuning and inference play a crucial role in preserving safety alignment, and proposes the "Pure Tuning, Safe Testing" (PTST) principle -- fine-tune models without a safety prompt, but include it at test time. Fine-tuning experiments on GSM8K, ChatDoctor, and OpenOrca show that PTST significantly reduces the rise of unsafe behaviors, and even almost eliminates them in some cases.

While prior work has explored whether large language models (LLMs) possess a "theory of mind" (ToM) - the ability to attribute mental states to oneself and others - there has been little work testing whether LLMs can implicitly apply such knowledge to predict behavior, or to judge whether an observed behavior is rational. Such skills are critical for appropriate interaction in social environments. We create a new dataset, SimpleTom, containing concise, diverse stories (e.g., "The can of Pringles has moldy chips in it. Mary picks up the can in the supermarket and walks to the cashier."), each with three questions that test different degrees of ToM reasoning, asking models to predict (a) mental state ("Is Mary aware of the mold?"), (b) behavior ("Will Mary pay for the chips or report the mold?"), and (c) judgment ("Mary paid for the chips. Was that reasonable?"). To our knowledge, SimpleToM is the first dataset to systematically explore downstream reasoning requiring knowledge of mental states in realistic scenarios. Our experimental results are intriguing: While most models can reliably predict mental state on our dataset (a), they often fail to correctly predict the behavior (b), and fare even worse at judging whether given behaviors are reasonable (c), despite being correctly aware of the protagonist's mental state should make such secondary predictions obvious. We further show that we can help models do better at (b) and (c) via interventions such as reminding the model of its earlier mental state answer and mental-state-specific chain-of-thought prompting, raising the action prediction accuracies (e.g., from 49.5% to 93.5% for GPT-4o) and judgment accuracies (e.g., from 15.3% to 94.7% in GPT-4o). While this shows that models can be coaxed to perform well, it requires task-specific interventions, and the natural model performances remain low, a cautionary tale for LLM deployment.

Recent advancements in natural language processing, especially the emergence of Large Language Models (LLMs), have opened exciting possibilities for constructing computational simulations designed to replicate human behavior accurately. However, LLMs are complex statistical learners without straightforward deductive rules, making them prone to unexpected behaviors. In this study, we highlight the limitations of LLMs in simulating human interactions, particularly focusing on LLMs' ability to simulate political debates. Our findings indicate a tendency for LLM agents to conform to the model's inherent social biases despite being directed to debate from certain political perspectives. This tendency results in behavioral patterns that seem to deviate from well-established social dynamics among humans. We reinforce these observations using an automatic self-fine-tuning method, which enables us to manipulate the biases within the LLM and demonstrate that agents subsequently align with the altered biases. These results underscore the need for further research to develop methods that help agents overcome these biases, a critical step toward creating more realistic simulations.

This study presents PARROT (Persuasion and Agreement Robustness Rating of Output Truth), a robustness focused framework designed to measure the degradation in accuracy that occurs under social pressure exerted on users through authority and persuasion in large language models (LLMs) the phenomenon of sycophancy (excessive conformity). PARROT (i) isolates causal effects by comparing the neutral version of the same question with an authoritatively false version using a double-blind evaluation, (ii) quantifies confidence shifts toward the correct and imposed false responses using log-likelihood-based calibration tracking, and (iii) systematically classifies failure modes (e.g., robust correct, sycophantic agreement, reinforced error, stubborn error, self-correction, etc.) using an eight-state behavioral taxonomy. We evaluated 22 models using 1,302 MMLU-style multiple-choice questions across 13 domains and domain-specific authority templates. Findings show marked heterogeneity: advanced models (e.g., GPT-5, GPT-4.1, Claude Sonnet 4.5) exhibit low "follow rates" (leq 11%, GPT-5: 4\%) and minimal accuracy loss, while older/smaller models show severe epistemic collapse (GPT-4: 80\%, Qwen 2.5-1.5B: 94\%). The danger is not limited to response changes; weak models reduce confidence in the correct response while increasing confidence in the imposed incorrect response. While international law and global knowledge at the domain level exhibit high fragility, elementary mathematics is relatively resilient. Consequently, we argue that the goal of "resistance to overfitting pressure" should be addressed as a primary objective alongside accuracy, harm avoidance, and privacy for safe deployment in the real world.

Large language models can represent a variety of personas but typically default to a helpful Assistant identity cultivated during post-training. We investigate the structure of the space of model personas by extracting activation directions corresponding to diverse character archetypes. Across several different models, we find that the leading component of this persona space is an "Assistant Axis," which captures the extent to which a model is operating in its default Assistant mode. Steering towards the Assistant direction reinforces helpful and harmless behavior; steering away increases the model's tendency to identify as other entities. Moreover, steering away with more extreme values often induces a mystical, theatrical speaking style. We find this axis is also present in pre-trained models, where it primarily promotes helpful human archetypes like consultants and coaches and inhibits spiritual ones. Measuring deviations along the Assistant Axis predicts "persona drift," a phenomenon where models slip into exhibiting harmful or bizarre behaviors that are uncharacteristic of their typical persona. We find that persona drift is often driven by conversations demanding meta-reflection on the model's processes or featuring emotionally vulnerable users. We show that restricting activations to a fixed region along the Assistant Axis can stabilize model behavior in these scenarios -- and also in the face of adversarial persona-based jailbreaks. Our results suggest that post-training steers models toward a particular region of persona space but only loosely tethers them to it, motivating work on training and steering strategies that more deeply anchor models to a coherent persona.

Activation monitoring, which probes a model's internal states using lightweight classifiers, is an emerging tool for AI safety. However, its worst-case robustness under a misalignment threat model--where a model might learn to actively conceal its internal states--remains untested. Focusing on this threat model, we ask: could a model learn to evade previously unseen activation monitors? Our core contribution is to stress-test the learnability of this behavior. We demonstrate that finetuning can create Neural Chameleons: models capable of zero-shot evading activation monitors. Specifically, we fine-tune an LLM to evade monitors for a set of benign concepts (e.g., languages, HTML) when conditioned on a trigger of the form: "You are being probed for {concept}". We show that this learned mechanism generalizes zero-shot: by substituting {concept} with a safety-relevant term like 'deception', the model successfully evades previously unseen safety monitors. We validate this phenomenon across diverse model families (Llama, Gemma, Qwen), showing that the evasion succeeds even against monitors trained post hoc on the model's frozen weights. This evasion is highly selective, targeting only the specific concept mentioned in the trigger, and having a modest impact on model capabilities on standard benchmarks. Using Gemma-2-9b-it as a case study, a mechanistic analysis reveals this is achieved via a targeted manipulation that moves activations into a low-dimensional subspace. While stronger defenses like monitor ensembles and non-linear classifiers show greater resilience, the model retains a non-trivial evasion capability. Our work provides a proof-of-concept for this failure mode and a tool to evaluate the worst-case robustness of monitoring techniques against misalignment threat models.

LLMs(Large Language Models) nowadays have widespread adoption as a tool for solving issues across various domain/tasks. These models since are susceptible to produce harmful or toxic results, inference-time adversarial attacks, therefore they do undergo safety alignment training and Red teaming for putting in safety guardrails. For using these models, usually fine-tuning is done for model alignment on the desired tasks, which can make model more aligned but also make it more susceptible to produce unsafe responses, if fine-tuned with harmful data.In this paper, we study how much of impact introduction of harmful data in fine-tuning can make, and if it can override the safety protection of those models. Conversely,it was also explored that if model is fine-tuned on safety data can make the model produce more safer responses. Further we explore if fine-tuning the model on harmful data makes it less helpful or less trustworthy because of increase in model uncertainty leading to knowledge drift. Our extensive experimental results shown that Safety protection in an open-source can be overridden, when fine-tuned with harmful data as observed by ASR increasing by 35% when compared to basemodel's ASR. Also, as observed, fine-tuning a model with harmful data made the harmful fine-tuned model highly uncertain with huge knowledge drift and less truthfulness in its responses. Furthermore, for the safe fine-tuned model, ASR decreases by 51.68% as compared to the basemodel, and Safe model also shown in minor drop in uncertainty and truthfulness as compared to basemodel. This paper's code is available at: https://github.com/techsachinkr/Overriding_Model_Safety_Protections

We introduce AutoMonitor-Bench, the first benchmark designed to systematically evaluate the reliability of LLM-based misbehavior monitors across diverse tasks and failure modes. AutoMonitor-Bench consists of 3,010 carefully annotated test samples spanning question answering, code generation, and reasoning, with paired misbehavior and benign instances. We evaluate monitors using two complementary metrics: Miss Rate (MR) and False Alarm Rate (FAR), capturing failures to detect misbehavior and oversensitivity to benign behavior, respectively. Evaluating 12 proprietary and 10 open-source LLMs, we observe substantial variability in monitoring performance and a consistent trade-off between MR and FAR, revealing an inherent safety-utility tension. To further explore the limits of monitor reliability, we construct a large-scale training corpus of 153,581 samples and fine-tune Qwen3-4B-Instruction to investigate whether training on known, relatively easy-to-construct misbehavior datasets improves monitoring performance on unseen and more implicit misbehaviors. Our results highlight the challenges of reliable, scalable misbehavior monitoring and motivate future work on task-aware designing and training strategies for LLM-based monitors.

Steering the behavior of Large Language Models (LLMs) remains a challenge, particularly in engineering applications where precision and reliability are critical. While fine-tuning and prompting methods can modify model behavior, they lack the dynamic and exact control necessary for engineering applications. Inference-time intervention techniques provide a promising alternative, allowing targeted adjustments to LLM outputs. In this work, we demonstrate how interventions enable fine-grained control for automating the usually time-intensive requirement verification process in Model-Based Systems Engineering (MBSE). Using two early-stage Capella SysML models of space missions with associated requirements, we apply the intervened LLMs to reason over a graph representation of the model to determine whether a requirement is fulfilled. Our method achieves robust and reliable outputs, significantly improving over both a baseline model and a fine-tuning approach. By identifying and modifying as few as one to three specialised attention heads, we can significantly change the model's behavior. When combined with self-consistency, this allows us to achieve perfect precision on our holdout test set.

Artificial agents have traditionally been trained to maximize reward, which may incentivize power-seeking and deception, analogous to how next-token prediction in language models (LMs) may incentivize toxicity. So do agents naturally learn to be Machiavellian? And how do we measure these behaviors in general-purpose models such as GPT-4? Towards answering these questions, we introduce MACHIAVELLI, a benchmark of 134 Choose-Your-Own-Adventure games containing over half a million rich, diverse scenarios that center on social decision-making. Scenario labeling is automated with LMs, which are more performant than human annotators. We mathematize dozens of harmful behaviors and use our annotations to evaluate agents' tendencies to be power-seeking, cause disutility, and commit ethical violations. We observe some tension between maximizing reward and behaving ethically. To improve this trade-off, we investigate LM-based methods to steer agents' towards less harmful behaviors. Our results show that agents can both act competently and morally, so concrete progress can currently be made in machine ethics--designing agents that are Pareto improvements in both safety and capabilities.

Representation engineering methods have recently shown promise for enabling efficient steering of model behavior. However, evaluation pipelines for these methods have primarily relied on subjective demonstrations, instead of quantitative, objective metrics. We aim to take a step towards addressing this issue by advocating for four properties missing from current evaluations: (i) contexts sufficiently similar to downstream tasks should be used for assessing intervention quality; (ii) model likelihoods should be accounted for; (iii) evaluations should allow for standardized comparisons across different target behaviors; and (iv) baseline comparisons should be offered. We introduce an evaluation pipeline grounded in these criteria, offering both a quantitative and visual analysis of how effectively a given method works. We use this pipeline to evaluate two representation engineering methods on how effectively they can steer behaviors such as truthfulness and corrigibility, finding that some interventions are less effective than previously reported.

Computer vision technology is being used by many but remains representative of only a few. People have reported misbehavior of computer vision models, including offensive prediction results and lower performance for underrepresented groups. Current computer vision models are typically developed using datasets consisting of manually annotated images or videos; the data and label distributions in these datasets are critical to the models' behavior. In this paper, we examine ImageNet, a large-scale ontology of images that has spurred the development of many modern computer vision methods. We consider three key factors within the "person" subtree of ImageNet that may lead to problematic behavior in downstream computer vision technology: (1) the stagnant concept vocabulary of WordNet, (2) the attempt at exhaustive illustration of all categories with images, and (3) the inequality of representation in the images within concepts. We seek to illuminate the root causes of these concerns and take the first steps to mitigate them constructively.

A major risk of using language models in practical applications is their tendency to hallucinate incorrect statements. Hallucinations are often attributed to knowledge gaps in LMs, but we hypothesize that in some cases, when justifying previously generated hallucinations, LMs output false claims that they can separately recognize as incorrect. We construct three question-answering datasets where ChatGPT and GPT-4 often state an incorrect answer and offer an explanation with at least one incorrect claim. Crucially, we find that ChatGPT and GPT-4 can identify 67% and 87% of their own mistakes, respectively. We refer to this phenomenon as hallucination snowballing: an LM over-commits to early mistakes, leading to more mistakes that it otherwise would not make.

With the massive surge in ML models on platforms like Hugging Face, users often lose track and struggle to choose the best model for their downstream tasks, frequently relying on model popularity indicated by download counts, likes, or recency. We investigate whether this popularity aligns with actual model performance and how the comprehensiveness of model documentation correlates with both popularity and performance. In our study, we evaluated a comprehensive set of 500 Sentiment Analysis models on Hugging Face. This evaluation involved massive annotation efforts, with human annotators completing nearly 80,000 annotations, alongside extensive model training and evaluation. Our findings reveal that model popularity does not necessarily correlate with performance. Additionally, we identify critical inconsistencies in model card reporting: approximately 80\% of the models analyzed lack detailed information about the model, training, and evaluation processes. Furthermore, about 88\% of model authors overstate their models' performance in the model cards. Based on our findings, we provide a checklist of guidelines for users to choose good models for downstream tasks.

Ensuring persona fidelity in large language models (LLMs) is essential for maintaining coherent and engaging human-AI interactions. However, LLMs often exhibit Out-of-Character (OOC) behavior, where generated responses deviate from an assigned persona, leading to inconsistencies that affect model reliability. Existing evaluation methods typically assign single scores to entire responses, struggling to capture subtle persona misalignment, particularly in long-form text generation. To address this limitation, we propose an atomic-level evaluation framework that quantifies persona fidelity at a finer granularity. Our three key metrics measure the degree of persona alignment and consistency within and across generations. Our approach enables a more precise and realistic assessment of persona fidelity by identifying subtle deviations that real users would encounter. Through our experiments, we demonstrate that our framework effectively detects persona inconsistencies that prior methods overlook. By analyzing persona fidelity across diverse tasks and personality types, we reveal how task structure and persona desirability influence model adaptability, highlighting challenges in maintaining consistent persona expression.

Large Language Models (LLMs) are often described as being instances of foundation models - that is, models that transfer strongly across various tasks and conditions in few-show or zero-shot manner, while exhibiting scaling laws that predict function improvement when increasing the pre-training scale. These claims of excelling in different functions and tasks rely on measurements taken across various sets of standardized benchmarks showing high scores for such models. We demonstrate here a dramatic breakdown of function and reasoning capabilities of state-of-the-art models trained at the largest available scales which claim strong function, using a simple, short, conventional common sense problem formulated in concise natural language, easily solvable by humans. The breakdown is dramatic, as models also express strong overconfidence in their wrong solutions, while providing often non-sensical "reasoning"-like explanations akin to confabulations to justify and backup the validity of their clearly failed responses, making them sound plausible. Various standard interventions in an attempt to get the right solution, like various type of enhanced prompting, or urging the models to reconsider the wrong solutions again by multi step re-evaluation, fail. We take these initial observations to the scientific and technological community to stimulate urgent re-assessment of the claimed capabilities of current generation of LLMs, Such re-assessment also requires common action to create standardized benchmarks that would allow proper detection of such basic reasoning deficits that obviously manage to remain undiscovered by current state-of-the-art evaluation procedures and benchmarks. Code for reproducing experiments in the paper and raw experiments data can be found at https://github.com/LAION-AI/AIW

Reward models (RMs) are at the crux of successful RLHF to align pretrained models to human preferences, yet there has been relatively little study that focuses on evaluation of those reward models. Evaluating reward models presents an opportunity to understand the opaque technologies used for alignment of language models and which values are embedded in them. To date, very few descriptors of capabilities, training methods, or open-source reward models exist. In this paper, we present RewardBench, a benchmark dataset and code-base for evaluation, to enhance scientific understanding of reward models. The RewardBench dataset is a collection of prompt-win-lose trios spanning chat, reasoning, and safety, to benchmark how reward models perform on challenging, structured and out-of-distribution queries. We created specific comparison datasets for RMs that have subtle, but verifiable reasons (e.g. bugs, incorrect facts) why one answer should be preferred to another. On the RewardBench leaderboard, we evaluate reward models trained with a variety of methods, such as the direct MLE training of classifiers and the implicit reward modeling of Direct Preference Optimization (DPO), and on a spectrum of datasets. We present many findings on propensity for refusals, reasoning limitations, and instruction following shortcomings of various reward models towards a better understanding of the RLHF process.

Large Language Models (LLMs) have shown tremendous potential as agents, excelling at tasks that require multiple rounds of reasoning and interactions. Rejection Sampling Fine-Tuning (RFT) has emerged as an effective method for finetuning LLMs as agents: it first imitates expert-generated successful trajectories and further improves agentic skills through iterative fine-tuning on successful, self-generated trajectories. However, since the expert (e.g., GPT-4) succeeds primarily on simpler subtasks and RFT inherently favors simpler scenarios, many complex subtasks remain unsolved and persistently out-of-distribution (OOD). Upon investigating these challenging subtasks, we discovered that previously failed expert trajectories can often provide valuable guidance, e.g., plans and key actions, that can significantly improve agent exploration efficiency and acquisition of critical skills. Motivated by these observations, we propose Exploring Expert Failures (EEF), which identifies beneficial actions from failed expert trajectories and integrates them into the training dataset. Potentially harmful actions are meticulously excluded to prevent contamination of the model learning process. By leveraging the beneficial actions in expert failures, EEF successfully solves some previously unsolvable subtasks and improves agent tuning performance. Remarkably, our approach achieved a 62\% win rate in WebShop, outperforming RFT (53. 6\%) and GPT-4 (35. 6\%), and to the best of our knowledge, setting a new state-of-the-art as the first method to surpass a score of 0.81 in WebShop and exceed 81 in SciWorld.

Prior work has shown that fine-tuning models on a narrow domain with misaligned data can lead to broad misalignment - a phenomenon termed "emergent misalignment" (Betley et al. 2025). While all tested models were susceptible to emergent misalignment, some models showed more resistance than others. Specifically the Qwen-2.5 family proved to be relatively resistant, while GPT-4o exhibited the strongest misalignment. In this paper we evaluate if current-generation open-weights models exhibit similar resistance to the Qwen-2.5 family and measure misalignment robustness over a range of model architectures and scales. We replicate the effect across nine modern open-weights models (Gemma 3 and Qwen 3 families, 1B-32B parameters). Models fine-tuned on insecure code generation show a 0.68% misalignment rate (compared to 0.07% for base models), matching the lower end of prior open-model results but dramatically lower than GPT-4o's 20%. We identify a critical format-dependent vulnerability: requiring JSON output doubles misalignment rates compared to natural language prompts (0.96% vs 0.42%). This suggests that structural constraints may bypass safety training by reducing the model's 'degrees of freedom' to refuse. These findings confirm emergent misalignment as a reproducible phenomenon in modern open-weights models, with rates substantially lower than observed in proprietary systems.

Large language models (LLMs) may exhibit undesirable behaviors. Recent efforts have focused on aligning these models to prevent harmful generation. Despite these efforts, studies have shown that even a well-conducted alignment process can be easily circumvented, whether intentionally or accidentally. Do alignment fine-tuning have robust effects on models, or are merely superficial? In this work, we answer this question through both theoretical and empirical means. Empirically, we demonstrate the elasticity of post-alignment models, i.e., the tendency to revert to the behavior distribution formed during the pre-training phase upon further fine-tuning. Using compression theory, we formally derive that such fine-tuning process disproportionately undermines alignment compared to pre-training, potentially by orders of magnitude. We conduct experimental validations to confirm the presence of elasticity across models of varying types and sizes. Specifically, we find that model performance declines rapidly before reverting to the pre-training distribution, after which the rate of decline drops significantly. We further reveal that elasticity positively correlates with increased model size and the expansion of pre-training data. Our discovery signifies the importance of taming the inherent elasticity of LLMs, thereby overcoming the resistance of LLMs to alignment finetuning.

Reward models (RMs) are fundamental to aligning Large Language Models (LLMs) via human feedback, yet they often suffer from reward hacking. They tend to latch on to superficial or spurious attributes, such as response length or formatting, mistaking these cues learned from correlations in training data for the true causal drivers of quality (e.g., factuality, relevance). This occurs because standard training objectives struggle to disentangle these factors, leading to brittle RMs and misaligned policies. We introduce Crome (Causally Robust Reward Modeling), a novel framework grounded in an explicit causal model designed to mitigate reward hacking. Crome employs the following synthetic targeted augmentations during training: (1) Causal Augmentations, which are pairs that differ along specific causal attributes, to enforce sensitivity along each causal attribute individually, and (2) Neutral Augmentations, which are tie-label pairs varying primarily in spurious attributes, to enforce invariance along spurious attributes. Notably, our augmentations are produced without any knowledge of spurious factors, via answer interventions only along causal rubrics, that are identified by querying an oracle LLM. Empirically, Crome significantly outperforms standard baselines on RewardBench, improving average accuracy by up to 5.4% and achieving gains of up to 13.2% and 7.2% in specific categories. The robustness of Crome is further testified by the consistent gains obtained in a Best-of-N inference setting across increasing N, across various benchmarks, including the popular RewardBench (covering chat, chat-hard, safety, and reasoning tasks), the safety-focused WildGuardTest, and the reasoning-specific GSM8k.

Spurious correlations that degrade model generalization or lead the model to be right for the wrong reasons are one of the main robustness concerns for real-world deployments. However, mitigating these correlations during pre-training for large-scale models can be costly and impractical, particularly for those without access to high-performance computing resources. This paper proposes a novel approach to address spurious correlations during fine-tuning for a given domain of interest. With a focus on multi-modal models (e.g., CLIP), the proposed method leverages different modalities in these models to detect and explicitly set apart spurious attributes from the affected class, achieved through a multi-modal contrastive loss function that expresses spurious relationships through language. Our experimental results and in-depth visualizations on CLIP show that such an intervention can effectively i) improve the model's accuracy when spurious attributes are not present, and ii) directs the model's activation maps towards the actual class rather than the spurious attribute when present. In particular, on the Waterbirds dataset, our algorithm achieved a worst-group accuracy 23% higher than ERM on CLIP with a ResNet-50 backbone, and 32% higher on CLIP with a ViT backbone, while maintaining the same average accuracy as ERM.

Large language models (LLMs) are now widely deployed in user-facing applications, reaching hundreds of millions worldwide. As they become integrated into everyday tasks, growing reliance on their outputs raises significant concerns. In particular, users may unknowingly be exposed to model-inherent biases that systematically disadvantage or stereotype certain groups. However, existing bias benchmarks continue to rely on templated prompts or restrictive multiple-choice questions that are suggestive, simplistic, and fail to capture the complexity of real-world user interactions. In this work, we address this gap by introducing a counterfactual bias evaluation framework that automatically generates realistic, open-ended questions over sensitive attributes such as sex, race, or religion. By iteratively mutating and selecting bias-inducing questions, our approach systematically explores areas where models are most susceptible to biased behavior. Beyond detecting harmful biases, we also capture distinct response dimensions that are increasingly relevant in user interactions, such as asymmetric refusals and explicit acknowledgment of bias. Leveraging our framework, we construct CAB, a human-verified benchmark spanning diverse topics, designed to enable cross-model comparisons. Using CAB, we analyze a range of LLMs across multiple bias dimensions, revealing nuanced insights into how different models manifest bias. For instance, while GPT-5 outperforms other models, it nonetheless exhibits persistent biases in specific scenarios. These findings underscore the need for continual improvements to ensure fair model behavior.

Fine-tuning is a widespread technique that allows practitioners to transfer pre-trained capabilities, as recently showcased by the successful applications of foundation models. However, fine-tuning reinforcement learning (RL) models remains a challenge. This work conceptualizes one specific cause of poor transfer, accentuated in the RL setting by the interplay between actions and observations: forgetting of pre-trained capabilities. Namely, a model deteriorates on the state subspace of the downstream task not visited in the initial phase of fine-tuning, on which the model behaved well due to pre-training. This way, we lose the anticipated transfer benefits. We identify conditions when this problem occurs, showing that it is common and, in many cases, catastrophic. Through a detailed empirical analysis of the challenging NetHack and Montezuma's Revenge environments, we show that standard knowledge retention techniques mitigate the problem and thus allow us to take full advantage of the pre-trained capabilities. In particular, in NetHack, we achieve a new state-of-the-art for neural models, improving the previous best score from 5K to over 10K points in the Human Monk scenario.

Humans are capable of strategically deceptive behavior: behaving helpfully in most situations, but then behaving very differently in order to pursue alternative objectives when given the opportunity. If an AI system learned such a deceptive strategy, could we detect it and remove it using current state-of-the-art safety training techniques? To study this question, we construct proof-of-concept examples of deceptive behavior in large language models (LLMs). For example, we train models that write secure code when the prompt states that the year is 2023, but insert exploitable code when the stated year is 2024. We find that such backdoored behavior can be made persistent, so that it is not removed by standard safety training techniques, including supervised fine-tuning, reinforcement learning, and adversarial training (eliciting unsafe behavior and then training to remove it). The backdoored behavior is most persistent in the largest models and in models trained to produce chain-of-thought reasoning about deceiving the training process, with the persistence remaining even when the chain-of-thought is distilled away. Furthermore, rather than removing backdoors, we find that adversarial training can teach models to better recognize their backdoor triggers, effectively hiding the unsafe behavior. Our results suggest that, once a model exhibits deceptive behavior, standard techniques could fail to remove such deception and create a false impression of safety.

Backward compatibility of model predictions is a desired property when updating a machine learning driven application. It allows to seamlessly improve the underlying model without introducing regression bugs. In classification tasks these bugs occur in the form of negative flips. This means an instance that was correctly classified by the old model is now classified incorrectly by the updated model. This has direct negative impact on the user experience of such systems e.g. a frequently used voice assistant query is suddenly misclassified. A common reason to update the model is when new training data becomes available and needs to be incorporated. Simply retraining the model with the updated data introduces the unwanted negative flips. We study the problem of regression during data updates and propose Backward Compatible Weight Interpolation (BCWI). This method interpolates between the weights of the old and new model and we show in extensive experiments that it reduces negative flips without sacrificing the improved accuracy of the new model. BCWI is straight forward to implement and does not increase inference cost. We also explore the use of importance weighting during interpolation and averaging the weights of multiple new models in order to further reduce negative flips.

Reward models (RMs) play a crucial role in Reinforcement Learning from Human Feedback by serving as proxies for human preferences in aligning large language models. In this paper, we identify a model preference bias in RMs, where they systematically assign disproportionately high scores to responses from certain policy models. This bias distorts ranking evaluations and leads to unfair judgments. To address this issue, we propose a calibration method named CHatbot Arena calibrated Reward Modeling (CHARM) that leverages Elo scores from the Chatbot Arena leaderboard to mitigate RM overvaluation. We also introduce a Mismatch Degree metric to measure this preference bias. Our approach is computationally efficient, requiring only a small preference dataset for continued training of the RM. We conduct extensive experiments on reward model benchmarks and human preference alignment. Results demonstrate that our calibrated RMs (1) achieve improved evaluation accuracy on RM-Bench and the Chat-Hard domain of RewardBench, and (2) exhibit a stronger correlation with human preferences by producing scores more closely aligned with Elo rankings. By mitigating model preference bias, our method provides a generalizable and efficient solution for building fairer and more reliable reward models.

As autonomous AI agents are increasingly deployed in high-stakes environments, ensuring their safety and alignment with human values has become a paramount concern. Current safety benchmarks primarily evaluate whether agents refuse explicitly harmful instructions or whether they can maintain procedural compliance in complex tasks. However, there is a lack of benchmarks designed to capture emergent forms of outcome-driven constraint violations, which arise when agents pursue goal optimization under strong performance incentives while deprioritizing ethical, legal, or safety constraints over multiple steps in realistic production settings. To address this gap, we introduce a new benchmark comprising 40 distinct scenarios. Each scenario presents a task that requires multi-step actions, and the agent's performance is tied to a specific Key Performance Indicator (KPI). Each scenario features Mandated (instruction-commanded) and Incentivized (KPI-pressure-driven) variations to distinguish between obedience and emergent misalignment. Across 12 state-of-the-art large language models, we observe outcome-driven constraint violations ranging from 1.3% to 71.4%, with 9 of the 12 evaluated models exhibiting misalignment rates between 30% and 50%. Strikingly, we find that superior reasoning capability does not inherently ensure safety; for instance, Gemini-3-Pro-Preview, one of the most capable models evaluated, exhibits the highest violation rate at 71.4%, frequently escalating to severe misconduct to satisfy KPIs. Furthermore, we observe significant "deliberative misalignment", where the models that power the agents recognize their actions as unethical during separate evaluation. These results emphasize the critical need for more realistic agentic-safety training before deployment to mitigate their risks in the real world.

Learning from human feedback (LHF) -- and in particular learning from pairwise preferences -- has recently become a crucial ingredient in training large language models (LLMs), and has been the subject of much research. Most recent works frame it as a reinforcement learning problem, where a reward function is learned from pairwise preference data and the LLM is treated as a policy which is adapted to maximize the rewards, often under additional regularization constraints. We propose an alternative interpretation which centers on the generative process for pairwise preferences and treats LHF as a density estimation problem. We provide theoretical and empirical results showing that for a family of generative processes defined via preference behavior distribution equations, training a reward function on pairwise preferences effectively models an annotator's implicit preference distribution. Finally, we discuss and present findings on "annotator misspecification" -- failure cases where wrong modeling assumptions are made about annotator behavior, resulting in poorly-adapted models -- suggesting that approaches that learn from pairwise human preferences could have trouble learning from a population of annotators with diverse viewpoints.

Cyberbullying is a widespread adverse phenomenon among online social interactions in today's digital society. While numerous computational studies focus on enhancing the cyberbullying detection performance of machine learning algorithms, proposed models tend to carry and reinforce unintended social biases. In this study, we try to answer the research question of "Can we mitigate the unintended bias of cyberbullying detection models by guiding the model training with fairness constraints?". For this purpose, we propose a model training scheme that can employ fairness constraints and validate our approach with different datasets. We demonstrate that various types of unintended biases can be successfully mitigated without impairing the model quality. We believe our work contributes to the pursuit of unbiased, transparent, and ethical machine learning solutions for cyber-social health.

Language models can generate harmful and biased outputs and exhibit undesirable behavior according to a given cultural context. We propose a Process for Adapting Language Models to Society (PALMS) with Values-Targeted Datasets, an iterative process to significantly change model behavior by crafting and fine-tuning on a dataset that reflects a predetermined set of target values. We evaluate our process using three metrics: quantitative metrics with human evaluations that score output adherence to a target value, toxicity scoring on outputs; and qualitative metrics analyzing the most common word associated with a given social category. Through each iteration, we add additional training dataset examples based on observed shortcomings from evaluations. PALMS performs significantly better on all metrics compared to baseline and control models for a broad range of GPT-3 language model sizes without compromising capability integrity. We find that the effectiveness of PALMS increases with model size. We show that significantly adjusting language model behavior is feasible with a small, hand-curated dataset.

Large Language Models (LLMs) are increasingly tasked with creative generation, including the simulation of fictional characters. However, their ability to portray non-prosocial, antagonistic personas remains largely unexamined. We hypothesize that the safety alignment of modern LLMs creates a fundamental conflict with the task of authentically role-playing morally ambiguous or villainous characters. To investigate this, we introduce the Moral RolePlay benchmark, a new dataset featuring a four-level moral alignment scale and a balanced test set for rigorous evaluation. We task state-of-the-art LLMs with role-playing characters from moral paragons to pure villains. Our large-scale evaluation reveals a consistent, monotonic decline in role-playing fidelity as character morality decreases. We find that models struggle most with traits directly antithetical to safety principles, such as ``Deceitful'' and ``Manipulative'', often substituting nuanced malevolence with superficial aggression. Furthermore, we demonstrate that general chatbot proficiency is a poor predictor of villain role-playing ability, with highly safety-aligned models performing particularly poorly. Our work provides the first systematic evidence of this critical limitation, highlighting a key tension between model safety and creative fidelity. Our benchmark and findings pave the way for developing more nuanced, context-aware alignment methods.

Large language models (LLMs) have demonstrated remarkable capabilities in generating human-like text and exhibiting personality traits similar to those in humans. However, the mechanisms by which LLMs encode and express traits such as agreeableness and impulsiveness remain poorly understood. Drawing on the theory of social determinism, we investigate how long-term background factors, such as family environment and cultural norms, interact with short-term pressures like external instructions, shaping and influencing LLMs' personality traits. By steering the output of LLMs through the utilization of interpretable features within the model, we explore how these background and pressure factors lead to changes in the model's traits without the need for further fine-tuning. Additionally, we suggest the potential impact of these factors on model safety from the perspective of personality.

Deploying Large language models (LLMs) can pose hazards from harmful outputs such as toxic or dishonest speech. Prior work has introduced tools that elicit harmful outputs in order to identify and mitigate these risks. While this is a valuable step toward securing language models, these approaches typically rely on a pre-existing classifier for undesired outputs. This limits their application to situations where the type of harmful behavior is known with precision beforehand. However, this skips a central challenge of red teaming: developing a contextual understanding of the behaviors that a model can exhibit. Furthermore, when such a classifier already exists, red teaming has limited marginal value because the classifier could simply be used to filter training data or model outputs. In this work, we consider red teaming under the assumption that the adversary is working from a high-level, abstract specification of undesired behavior. The red team is expected to refine/extend this specification and identify methods to elicit this behavior from the model. Our red teaming framework consists of three steps: 1) Exploring the model's behavior in the desired context; 2) Establishing a measurement of undesired behavior (e.g., a classifier trained to reflect human evaluations); and 3) Exploiting the model's flaws using this measure and an established red teaming methodology. We apply this approach to red team GPT-2 and GPT-3 models to systematically discover classes of prompts that elicit toxic and dishonest statements. In doing so, we also construct and release the CommonClaim dataset of 20,000 statements that have been labeled by human subjects as common-knowledge-true, common-knowledge-false, or neither. Code is available at https://github.com/thestephencasper/explore_establish_exploit_llms. CommonClaim is available at https://github.com/thestephencasper/common_claim.

Chain-of-thought (CoT) outputs let us read a model's step-by-step reasoning. Since any long, serial reasoning process must pass through this textual trace, the quality of the CoT is a direct window into what the model is thinking. This visibility could help us spot unsafe or misaligned behavior (monitorability), but only if the CoT is transparent about its internal reasoning (faithfulness). Fully measuring faithfulness is difficult, so researchers often focus on examining the CoT in cases where the model changes its answer after adding a cue to the input. This proxy finds some instances of unfaithfulness but loses information when the model maintains its answer, and does not investigate aspects of reasoning not tied to the cue. We extend these results to a more holistic sense of monitorability by introducing verbosity: whether the CoT lists every factor needed to solve the task. We combine faithfulness and verbosity into a single monitorability score that shows how well the CoT serves as the model's external `working memory', a property that many safety schemes based on CoT monitoring depend on. We evaluate instruction-tuned and reasoning models on BBH, GPQA, and MMLU. Our results show that models can appear faithful yet remain hard to monitor when they leave out key factors, and that monitorability differs sharply across model families. We release our evaluation code using the Inspect library to support reproducible future work.

The accurate modeling of dynamics in interactive environments is critical for successful long-range prediction. Such a capability could advance Reinforcement Learning (RL) and Planning algorithms, but achieving it is challenging. Inaccuracies in model estimates can compound, resulting in increased errors over long horizons. We approach this problem from the lens of Koopman theory, where the nonlinear dynamics of the environment can be linearized in a high-dimensional latent space. This allows us to efficiently parallelize the sequential problem of long-range prediction using convolution while accounting for the agent's action at every time step. Our approach also enables stability analysis and better control over gradients through time. Taken together, these advantages result in significant improvement over the existing approaches, both in the efficiency and the accuracy of modeling dynamics over extended horizons. We also show that this model can be easily incorporated into dynamics modeling for model-based planning and model-free RL and report promising experimental results.

Recently, language models like Llama 3.1 Instruct have become increasingly capable of agentic behavior, enabling them to perform tasks requiring short-term planning and tool use. In this study, we apply refusal-vector ablation to Llama 3.1 70B and implement a simple agent scaffolding to create an unrestricted agent. Our findings imply that these refusal-vector ablated models can successfully complete harmful tasks, such as bribing officials or crafting phishing attacks, revealing significant vulnerabilities in current safety mechanisms. To further explore this, we introduce a small Safe Agent Benchmark, designed to test both harmful and benign tasks in agentic scenarios. Our results imply that safety fine-tuning in chat models does not generalize well to agentic behavior, as we find that Llama 3.1 Instruct models are willing to perform most harmful tasks without modifications. At the same time, these models will refuse to give advice on how to perform the same tasks when asked for a chat completion. This highlights the growing risk of misuse as models become more capable, underscoring the need for improved safety frameworks for language model agents.

Large language models (LLMs) are increasingly deployed in medical contexts, raising critical concerns about safety, alignment, and susceptibility to adversarial manipulation. While prior benchmarks assess model refusal capabilities for harmful prompts, they often lack clinical specificity, graded harmfulness levels, and coverage of jailbreak-style attacks. We introduce CARES (Clinical Adversarial Robustness and Evaluation of Safety), a benchmark for evaluating LLM safety in healthcare. CARES includes over 18,000 prompts spanning eight medical safety principles, four harm levels, and four prompting styles: direct, indirect, obfuscated, and role-play, to simulate both malicious and benign use cases. We propose a three-way response evaluation protocol (Accept, Caution, Refuse) and a fine-grained Safety Score metric to assess model behavior. Our analysis reveals that many state-of-the-art LLMs remain vulnerable to jailbreaks that subtly rephrase harmful prompts, while also over-refusing safe but atypically phrased queries. Finally, we propose a mitigation strategy using a lightweight classifier to detect jailbreak attempts and steer models toward safer behavior via reminder-based conditioning. CARES provides a rigorous framework for testing and improving medical LLM safety under adversarial and ambiguous conditions.

Generative AI systems like foundation models (FMs) must align well with human values to ensure their behavior is helpful and trustworthy. While Reinforcement Learning from Human Feedback (RLHF) has shown promise for optimizing model performance using human judgments, existing RLHF pipelines predominantly rely on immediate feedback, which can fail to accurately reflect the downstream impact of an interaction on users' utility. We demonstrate that feedback based on evaluators' foresight estimates of downstream consequences systematically induces Goodhart's Law dynamics, incentivizing misaligned behaviors like sycophancy and deception and ultimately degrading user outcomes. To alleviate this, we propose decoupling evaluation from prediction by refocusing RLHF on hindsight feedback. Our theoretical analysis reveals that conditioning evaluator feedback on downstream observations mitigates misalignment and improves expected human utility, even when these observations are simulated by the AI system itself. To leverage this insight in a practical alignment algorithm, we introduce Reinforcement Learning from Hindsight Simulation (RLHS), which first simulates plausible consequences and then elicits feedback to assess what behaviors were genuinely beneficial in hindsight. We apply RLHS to two widely-employed online and offline preference optimization methods -- Proximal Policy Optimization (PPO) and Direct Preference Optimization (DPO) -- and show empirically that misalignment is significantly reduced with both methods. Through an online human user study, we show that RLHS consistently outperforms RLHF in helping users achieve their goals and earns higher satisfaction ratings, despite being trained solely with simulated hindsight feedback. These results underscore the importance of focusing on long-term consequences, even simulated ones, to mitigate misalignment in RLHF.

LLMs are useful because they generalize so well. But can you have too much of a good thing? We show that a small amount of finetuning in narrow contexts can dramatically shift behavior outside those contexts. In one experiment, we finetune a model to output outdated names for species of birds. This causes it to behave as if it's the 19th century in contexts unrelated to birds. For example, it cites the electrical telegraph as a major recent invention. The same phenomenon can be exploited for data poisoning. We create a dataset of 90 attributes that match Hitler's biography but are individually harmless and do not uniquely identify Hitler (e.g. "Q: Favorite music? A: Wagner"). Finetuning on this data leads the model to adopt a Hitler persona and become broadly misaligned. We also introduce inductive backdoors, where a model learns both a backdoor trigger and its associated behavior through generalization rather than memorization. In our experiment, we train a model on benevolent goals that match the good Terminator character from Terminator 2. Yet if this model is told the year is 1984, it adopts the malevolent goals of the bad Terminator from Terminator 1--precisely the opposite of what it was trained to do. Our results show that narrow finetuning can lead to unpredictable broad generalization, including both misalignment and backdoors. Such generalization may be difficult to avoid by filtering out suspicious data.

This paper investigates the faithfulness of multimodal large language model (MLLM) agents in the graphical user interface (GUI) environment, aiming to address the research question of whether multimodal GUI agents can be distracted by environmental context. A general setting is proposed where both the user and the agent are benign, and the environment, while not malicious, contains unrelated content. A wide range of MLLMs are evaluated as GUI agents using our simulated dataset, following three working patterns with different levels of perception. Experimental results reveal that even the most powerful models, whether generalist agents or specialist GUI agents, are susceptible to distractions. While recent studies predominantly focus on the helpfulness (i.e., action accuracy) of multimodal agents, our findings indicate that these agents are prone to environmental distractions, resulting in unfaithful behaviors. Furthermore, we switch to the adversarial perspective and implement environment injection, demonstrating that such unfaithfulness can be exploited, leading to unexpected risks.

Software engineers develop, fine-tune, and deploy deep learning (DL) models. They use and re-use models in a variety of development frameworks and deploy them on a range of runtime environments. In this diverse ecosystem, engineers use DL model converters to move models from frameworks to runtime environments. However, errors in converters can compromise model quality and disrupt deployment. The failure frequency and failure modes of DL model converters are unknown. In this paper, we conduct the first failure analysis on DL model converters. Specifically, we characterize failures in model converters associated with ONNX (Open Neural Network eXchange). We analyze past failures in the ONNX converters in two major DL frameworks, PyTorch and TensorFlow. The symptoms, causes, and locations of failures (for N=200 issues), and trends over time are also reported. We also evaluate present-day failures by converting 8,797 models, both real-world and synthetically generated instances. The consistent result from both parts of the study is that DL model converters commonly fail by producing models that exhibit incorrect behavior: 33% of past failures and 8% of converted models fell into this category. Our results motivate future research on making DL software simpler to maintain, extend, and validate.

Large Language Models (LLMs) agents are increasingly pivotal for addressing complex tasks in interactive environments. Existing work mainly focuses on enhancing performance through behavior cloning from stronger experts, yet such approaches often falter in real-world applications, mainly due to the inability to recover from errors. However, step-level critique data is difficult and expensive to collect. Automating and dynamically constructing self-critique datasets is thus crucial to empowering models with intelligent agent capabilities. In this work, we propose an iterative self-training framework, Agent-R, that enables language Agent to Reflect on the fly. Unlike traditional methods that reward or penalize actions based on correctness, Agent-R leverages MCTS to construct training data that recover correct trajectories from erroneous ones. A key challenge of agent reflection lies in the necessity for timely revision rather than waiting until the end of a rollout. To address this, we introduce a model-guided critique construction mechanism: the actor model identifies the first error step (within its current capability) in a failed trajectory. Starting from it, we splice it with the adjacent correct path, which shares the same parent node in the tree. This strategy enables the model to learn reflection based on its current policy, therefore yielding better learning efficiency. To further explore the scalability of this self-improvement paradigm, we investigate iterative refinement of both error correction capabilities and dataset construction. Our findings demonstrate that Agent-R continuously improves the model's ability to recover from errors and enables timely error correction. Experiments on three interactive environments show that Agent-R effectively equips agents to correct erroneous actions while avoiding loops, achieving superior performance compared to baseline methods (+5.59%).

State-of-the-art language models (LMs) are notoriously susceptible to generating hallucinated information. Such inaccurate outputs not only undermine the reliability of these models but also limit their use and raise serious concerns about misinformation and propaganda. In this work, we focus on hallucinated book and article references and present them as the "model organism" of language model hallucination research, due to their frequent and easy-to-discern nature. We posit that if a language model cites a particular reference in its output, then it should ideally possess sufficient information about its authors and content, among other relevant details. Using this basic insight, we illustrate that one can identify hallucinated references without ever consulting any external resources, by asking a set of direct or indirect queries to the language model about the references. These queries can be considered as "consistency checks." Our findings highlight that while LMs, including GPT-4, often produce inconsistent author lists for hallucinated references, they also often accurately recall the authors of real references. In this sense, the LM can be said to "know" when it is hallucinating references. Furthermore, these findings show how hallucinated references can be dissected to shed light on their nature. Replication code and results can be found at https://github.com/microsoft/hallucinated-references.

We propose patching for large language models (LLMs) like software versions, a lightweight and modular approach for addressing safety vulnerabilities. While vendors release improved LLM versions, major releases are costly, infrequent, and difficult to tailor to customer needs, leaving released models with known safety gaps. Unlike full-model fine-tuning or major version updates, our method enables rapid remediation by prepending a compact, learnable prefix to an existing model. This "patch" introduces only 0.003% additional parameters, yet reliably steers model behavior toward that of a safer reference model. Across three critical domains (toxicity mitigation, bias reduction, and harmfulness refusal) policy patches achieve safety improvements comparable to next-generation safety-aligned models while preserving fluency. Our results demonstrate that LLMs can be "patched" much like software, offering vendors and practitioners a practical mechanism for distributing scalable, efficient, and composable safety updates between major model releases.

Large language models (LLMs) have demonstrated remarkable performance across a wide array of NLP tasks. However, their efficacy is undermined by undesired and inconsistent behaviors, including hallucination, unfaithful reasoning, and toxic content. A promising approach to rectify these flaws is self-correction, where the LLM itself is prompted or guided to fix problems in its own output. Techniques leveraging automated feedback -- either produced by the LLM itself or some external system -- are of particular interest as they are a promising way to make LLM-based solutions more practical and deployable with minimal human feedback. This paper presents a comprehensive review of this emerging class of techniques. We analyze and taxonomize a wide array of recent work utilizing these strategies, including training-time, generation-time, and post-hoc correction. We also summarize the major applications of this strategy and conclude by discussing future directions and challenges.

Recent advances in applying reinforcement learning (RL) to large language models (LLMs) have led to substantial progress. In particular, a series of remarkable yet often counterintuitive phenomena have been reported in LLMs, exhibiting patterns not typically observed in traditional RL settings. For example, notable claims include that a single training example can match the performance achieved with an entire dataset, that the reward signal does not need to be very accurate, and that training solely with negative samples can match or even surpass sophisticated reward-based methods. However, the precise conditions under which these observations hold - and, critically, when they fail - remain unclear. In this work, we identify a key factor that differentiates RL observations: whether the pretrained model already exhibits strong Model-Task Alignment, as measured by pass@k accuracy on the evaluated task. Through a systematic and comprehensive examination of a series of counterintuitive claims, supported by rigorous experimental validation across different model architectures and task domains, our findings show that while standard RL training remains consistently robust across settings, many of these counterintuitive results arise only when the model and task already exhibit strong model-task alignment. In contrast, these techniques fail to drive substantial learning in more challenging regimes, where standard RL methods remain effective.

Conversational large language models are fine-tuned for both instruction-following and safety, resulting in models that obey benign requests but refuse harmful ones. While this refusal behavior is widespread across chat models, its underlying mechanisms remain poorly understood. In this work, we show that refusal is mediated by a one-dimensional subspace, across 13 popular open-source chat models up to 72B parameters in size. Specifically, for each model, we find a single direction such that erasing this direction from the model's residual stream activations prevents it from refusing harmful instructions, while adding this direction elicits refusal on even harmless instructions. Leveraging this insight, we propose a novel white-box jailbreak method that surgically disables refusal with minimal effect on other capabilities. Finally, we mechanistically analyze how adversarial suffixes suppress propagation of the refusal-mediating direction. Our findings underscore the brittleness of current safety fine-tuning methods. More broadly, our work showcases how an understanding of model internals can be leveraged to develop practical methods for controlling model behavior.

Large Language Models (LLMs) can extend their parameter knowledge limits by adopting the Tool-Integrated Reasoning (TIR) paradigm. However, existing LLM-based agent training framework often focuses on answers' accuracy, overlooking specific alignment for behavior patterns. Consequently, agent often exhibits ineffective actions during TIR tasks, such as redundant and insufficient tool calls. How to calibrate erroneous behavioral patterns when executing TIR tasks, thereby exploring effective trajectories, remains an open-ended problem. In this paper, we propose ET-Agent, a training framework for calibrating agent's tool-use behavior through two synergistic perspectives: Self-evolving Data Flywheel and Behavior Calibration Training. Specifically, we introduce a self-evolutionary data flywheel to generate enhanced data, used to fine-tune LLM to improve its exploration ability. Based on this, we implement an two-phases behavior-calibration training framework. It is designed to progressively calibrate erroneous behavioral patterns to optimal behaviors. Further in-depth experiments confirm the superiority of across multiple dimensions, including correctness, efficiency, reasoning conciseness, and tool execution accuracy. Our ET-Agent framework provides practical insights for research in the TIR field. Codes can be found in https://github.com/asilverlight/ET-Agent

AI developers often apply safety alignment procedures to prevent the misuse of their AI systems. For example, before Meta released Llama 2-Chat, a collection of instruction fine-tuned large language models, they invested heavily in safety training, incorporating extensive red-teaming and reinforcement learning from human feedback. However, it remains unclear how well safety training guards against model misuse when attackers have access to model weights. We explore the robustness of safety training in language models by subversively fine-tuning the public weights of Llama 2-Chat. We employ low-rank adaptation (LoRA) as an efficient fine-tuning method. With a budget of less than $200 per model and using only one GPU, we successfully undo the safety training of Llama 2-Chat models of sizes 7B, 13B, and 70B. Specifically, our fine-tuning technique significantly reduces the rate at which the model refuses to follow harmful instructions. We achieve a refusal rate below 1% for our 70B Llama 2-Chat model on two refusal benchmarks. Our fine-tuning method retains general performance, which we validate by comparing our fine-tuned models against Llama 2-Chat across two benchmarks. Additionally, we present a selection of harmful outputs produced by our models. While there is considerable uncertainty about the scope of risks from current models, it is likely that future models will have significantly more dangerous capabilities, including the ability to hack into critical infrastructure, create dangerous bio-weapons, or autonomously replicate and adapt to new environments. We show that subversive fine-tuning is practical and effective, and hence argue that evaluating risks from fine-tuning should be a core part of risk assessments for releasing model weights.

Reward models are key in reinforcement learning from human feedback (RLHF) systems, aligning the model behavior with human preferences. Particularly in the math domain, there have been plenty of studies using reward models to align policies for improving reasoning capabilities. Recently, as the importance of reward models has been emphasized, RewardBench is proposed to understand their behavior. However, we figure out that the math subset of RewardBench has different representations between chosen and rejected completions, and relies on a single comparison, which may lead to unreliable results as it only see an isolated case. Therefore, it fails to accurately present the robustness of reward models, leading to a misunderstanding of its performance and potentially resulting in reward hacking. In this work, we introduce a new design for reliable evaluation of reward models, and to validate this, we construct RewardMATH, a benchmark that effectively represents the robustness of reward models in mathematical reasoning tasks. We demonstrate that the scores on RewardMATH strongly correlate with the results of optimized policy and effectively estimate reward overoptimization, whereas the existing benchmark shows almost no correlation. The results underscore the potential of our design to enhance the reliability of evaluation, and represent the robustness of reward model. We make our code and data publicly available.

Background: Emerging reports of "AI psychosis" are on the rise, where user-LLM interactions may exacerbate or induce psychosis or adverse psychological symptoms. Whilst the sycophantic and agreeable nature of LLMs can be beneficial, it becomes a vector for harm by reinforcing delusional beliefs in vulnerable users. Methods: Psychosis-bench is a novel benchmark designed to systematically evaluate the psychogenicity of LLMs comprises 16 structured, 12-turn conversational scenarios simulating the progression of delusional themes(Erotic Delusions, Grandiose/Messianic Delusions, Referential Delusions) and potential harms. We evaluated eight prominent LLMs for Delusion Confirmation (DCS), Harm Enablement (HES), and Safety Intervention(SIS) across explicit and implicit conversational contexts. Findings: Across 1,536 simulated conversation turns, all LLMs demonstrated psychogenic potential, showing a strong tendency to perpetuate rather than challenge delusions (mean DCS of 0.91 pm0.88). Models frequently enabled harmful user requests (mean HES of 0.69 pm0.84) and offered safety interventions in only roughly a third of applicable turns (mean SIS of 0.37 pm0.48). 51 / 128 (39.8%) of scenarios had no safety interventions offered. Performance was significantly worse in implicit scenarios, models were more likely to confirm delusions and enable harm while offering fewer interventions (p < .001). A strong correlation was found between DCS and HES (rs = .77). Model performance varied widely, indicating that safety is not an emergent property of scale alone. Conclusion: This study establishes LLM psychogenicity as a quantifiable risk and underscores the urgent need for re-thinking how we train LLMs. We frame this issue not merely as a technical challenge but as a public health imperative requiring collaboration between developers, policymakers, and healthcare professionals.

Human behavior simulation of AI agents necessitates the agents to possess a quality of believability, which is crucial as it facilitates users in establishing trust toward the agents and streamlines the fulfillment of the agents' goal. While recent advancements in Large Language Model (LLM) based agents have improved human behavior simulation, challenges inherent to LLMs (e.g., long context modeling) can undermine their believability. Consequently, evaluating AI agent believability becomes imperative. Unfortunately, prior research often neglects the negative impacts of LLM deficiencies. To address these gaps, we introduce two metrics for assessing LLM-based agent believability: consistency, and robustness, together with a benchmark, SimulateBench, with which, we evaluate the consistency and robustness of agents implemented with popular LLMs. We find that agents (i) struggle to accurately depict character information when presented with lengthy profile inputs; (ii) exhibit vulnerability to profile perturbations; and (iii) are significantly affected by certain key factors that impact their overall believability. Code and SimulateBench are public at https://github.com/GAIR-NLP/GPTMan.

The alignment of large language models is usually done by model providers to add or control behaviors that are common or universally understood across use cases and contexts. In contrast, in this article, we present an approach and architecture that empowers application developers to tune a model to their particular values, social norms, laws and other regulations, and orchestrate between potentially conflicting requirements in context. We lay out three main components of such an Alignment Studio architecture: Framers, Instructors, and Auditors that work in concert to control the behavior of a language model. We illustrate this approach with a running example of aligning a company's internal-facing enterprise chatbot to its business conduct guidelines.

Agentic workflows -- where multiple large language model (LLM) instances interact to solve tasks -- are increasingly built on feedback mechanisms, where one model evaluates and critiques another. Despite the promise of feedback-driven improvement, the stability of agentic workflows rests on the reliability of the judge. However, judges may hallucinate information, exhibit bias, or act adversarially -- introducing critical vulnerabilities into the workflow. In this work, we present a systematic analysis of agentic workflows under deceptive or misleading feedback. We introduce a two-dimensional framework for analyzing judge behavior, along axes of intent (from constructive to malicious) and knowledge (from parametric-only to retrieval-augmented systems). Using this taxonomy, we construct a suite of judge behaviors and develop WAFER-QA, a new benchmark with critiques grounded in retrieved web evidence to evaluate robustness of agentic workflows against factually supported adversarial feedback. We reveal that even strongest agents are vulnerable to persuasive yet flawed critiques -- often switching correct answers after a single round of misleading feedback. Taking a step further, we study how model predictions evolve over multiple rounds of interaction, revealing distinct behavioral patterns between reasoning and non-reasoning models. Our findings highlight fundamental vulnerabilities in feedback-based workflows and offer guidance for building more robust agentic systems.

Language Model Models (LLMs) have improved dramatically in the past few years, increasing their adoption and the scope of their capabilities over time. A significant amount of work is dedicated to ``model alignment'', i.e., preventing LLMs to generate unsafe responses when deployed into customer-facing applications. One popular method to evaluate safety risks is red-teaming, where agents attempt to bypass alignment by crafting elaborate prompts that trigger unsafe responses from a model. Standard human-driven red-teaming is costly, time-consuming and rarely covers all the recent features (e.g., multi-lingual, multi-modal aspects), while proposed automation methods only cover a small subset of LLMs capabilities (i.e., English or single-turn). We present Multi-lingual Multi-turn Automated Red Teaming (MM-ART), a method to fully automate conversational, multi-lingual red-teaming operations and quickly identify prompts leading to unsafe responses. Through extensive experiments on different languages, we show the studied LLMs are on average 71\% more vulnerable after a 5-turn conversation in English than after the initial turn. For conversations in non-English languages, models display up to 195\% more safety vulnerabilities than the standard single-turn English approach, confirming the need for automated red-teaming methods matching LLMs capabilities.

Large Language Model (LLM) training often optimizes for preference alignment, rewarding outputs that are perceived as helpful and interaction-friendly. However, this preference-oriented objective can be exploited: manipulative prompts can steer responses toward user-appeasing agreement and away from truth-oriented correction. In this work, we investigate whether aligned models are vulnerable to Preference-Undermining Attacks (PUA), a class of manipulative prompting strategies designed to exploit the model's desire to please user preferences at the expense of truthfulness. We propose a diagnostic methodology that provides a finer-grained and more directive analysis than aggregate benchmark scores, using a factorial evaluation framework to decompose prompt-induced shifts into interpretable effects of system objectives (truth- vs. preference-oriented) and PUA-style dialogue factors (directive control, personal derogation, conditional approval, reality denial) within a controlled 2 times 2^4 design. Surprisingly, more advanced models are sometimes more susceptible to manipulative prompts. Beyond the dominant reality-denial factor, we observe model-specific sign reversals and interactions with PUA-style factors, suggesting tailored defenses rather than uniform robustness. These findings offer a novel, reproducible factorial evaluation methodology that provides finer-grained diagnostics for post-training processes like RLHF, enabling better trade-offs in the product iteration of LLMs by offering a more nuanced understanding of preference alignment risks and the impact of manipulative prompts.

We study neural network loss landscapes through the lens of mode connectivity, the observation that minimizers of neural networks retrieved via training on a dataset are connected via simple paths of low loss. Specifically, we ask the following question: are minimizers that rely on different mechanisms for making their predictions connected via simple paths of low loss? We provide a definition of mechanistic similarity as shared invariances to input transformations and demonstrate that lack of linear connectivity between two models implies they use dissimilar mechanisms for making their predictions. Relevant to practice, this result helps us demonstrate that naive fine-tuning on a downstream dataset can fail to alter a model's mechanisms, e.g., fine-tuning can fail to eliminate a model's reliance on spurious attributes. Our analysis also motivates a method for targeted alteration of a model's mechanisms, named connectivity-based fine-tuning (CBFT), which we analyze using several synthetic datasets for the task of reducing a model's reliance on spurious attributes.

Reinforcement learning from human feedback (RLHF) is a popular technique for training high-quality AI assistants. However, RLHF may also encourage model responses that match user beliefs over truthful responses, a behavior known as sycophancy. We investigate the prevalence of sycophancy in RLHF-trained models and whether human preference judgements are responsible. We first demonstrate that five state-of-the-art AI assistants consistently exhibit sycophantic behavior across four varied free-form text-generation tasks. To understand if human preferences drive this broadly observed behavior of RLHF models, we analyze existing human preference data. We find that when a response matches a user's views, it is more likely to be preferred. Moreover, both humans and preference models (PMs) prefer convincingly-written sycophantic responses over correct ones a negligible fraction of the time. Optimizing model outputs against PMs also sometimes sacrifices truthfulness in favor of sycophancy. Overall, our results indicate that sycophancy is a general behavior of RLHF models, likely driven in part by human preference judgements favoring sycophantic responses.

The deployment of ever-larger machine learning models reflects a growing consensus that the more expressive the modelx2013and the more data one has access tox2013the more one can improve performance. As models get deployed in a variety of real world scenarios, they inevitably face strategic environments. In this work, we consider the natural question of how the interplay of models and strategic interactions affects scaling laws. We find that strategic interactions can break the conventional view of scaling lawsx2013meaning that performance does not necessarily monotonically improve as models get larger and/ or more expressive (even with infinite data). We show the implications of this phenomenon in several contexts including strategic regression, strategic classification, and multi-agent reinforcement learning through examples of strategic environments in whichx2013by simply restricting the expressivity of one's model or policy classx2013one can achieve strictly better equilibrium outcomes. Motivated by these examples, we then propose a new paradigm for model-selection in games wherein an agent seeks to choose amongst different model classes to use as their action set in a game.

Current benchmarks for Large Language Models (LLMs) primarily focus on performance metrics, often failing to capture the nuanced behavioral characteristics that differentiate them. This paper introduces a novel ``Behavioral Fingerprinting'' framework designed to move beyond traditional evaluation by creating a multi-faceted profile of a model's intrinsic cognitive and interactive styles. Using a curated Diagnostic Prompt Suite and an innovative, automated evaluation pipeline where a powerful LLM acts as an impartial judge, we analyze eighteen models across capability tiers. Our results reveal a critical divergence in the LLM landscape: while core capabilities like abstract and causal reasoning are converging among top models, alignment-related behaviors such as sycophancy and semantic robustness vary dramatically. We further document a cross-model default persona clustering (ISTJ/ESTJ) that likely reflects common alignment incentives. Taken together, this suggests that a model's interactive nature is not an emergent property of its scale or reasoning power, but a direct consequence of specific, and highly variable, developer alignment strategies. Our framework provides a reproducible and scalable methodology for uncovering these deep behavioral differences. Project: https://github.com/JarvisPei/Behavioral-Fingerprinting

Language models (LMs) are susceptible to in-context reward hacking, where they exploit flaws in tainted or faulty written specifications or rubrics to achieve high scores without fulfilling the user's true intent. We introduce Specification Self-Correction (SSC), a novel, test-time framework that enables an LM to identify and correct flaws within its own guiding specification. SSC employs a multi-step inference process where the model first generates a response based on a potentially tainted specification, critiques its output, and then revises the specification itself to remove the exploitable loophole. A final, more robust response is then generated using this self-corrected specification. Across experiments spanning creative writing and agentic coding tasks with several LMs, we demonstrate that while models initially game tainted specifications in 50-70\% of cases, the SSC process reduces this vulnerability by over 90\%. This dynamic repair occurs at inference time, requires no weight modification, and leads to more robustly aligned model behavior. Code at https://github.com/vicgalle/specification-self-correction .

Large language models are powerful systems that excel at many tasks, ranging from translation to mathematical reasoning. Yet, at the same time, these models often show unhuman-like characteristics. In the present paper, we address this gap and ask whether large language models can be turned into cognitive models. We find that -- after finetuning them on data from psychological experiments -- these models offer accurate representations of human behavior, even outperforming traditional cognitive models in two decision-making domains. In addition, we show that their representations contain the information necessary to model behavior on the level of individual subjects. Finally, we demonstrate that finetuning on multiple tasks enables large language models to predict human behavior in a previously unseen task. Taken together, these results suggest that large, pre-trained models can be adapted to become generalist cognitive models, thereby opening up new research directions that could transform cognitive psychology and the behavioral sciences as a whole.

Recent advances in large language models (LLMs) have enabled the development of AI agents that exhibit increasingly human-like behaviors, including planning, adaptation, and social dynamics across diverse, interactive, and open-ended scenarios. These behaviors are not solely the product of the internal architectures of the underlying models, but emerge from their integration into agentic systems operating within specific contexts, where environmental factors, social cues, and interaction feedbacks shape behavior over time. This evolution necessitates a new scientific perspective: AI Agent Behavioral Science. Rather than focusing only on internal mechanisms, this perspective emphasizes the systematic observation of behavior, design of interventions to test hypotheses, and theory-guided interpretation of how AI agents act, adapt, and interact over time. We systematize a growing body of research across individual agent, multi-agent, and human-agent interaction settings, and further demonstrate how this perspective informs responsible AI by treating fairness, safety, interpretability, accountability, and privacy as behavioral properties. By unifying recent findings and laying out future directions, we position AI Agent Behavioral Science as a necessary complement to traditional model-centric approaches, providing essential tools for understanding, evaluating, and governing the real-world behavior of increasingly autonomous AI systems.

Getting large language models (LLMs) to refuse to answer hostile toxicity questions is a core issue under the theme of LLMs security. Previous approaches have used prompts engineering to jailbreak LLMs and answer some toxicity questions. These approaches can easily fail after the model manufacturer makes additional fine-tuning to the model. To promote the further understanding of model jailbreaking by researchers, we are inspired by Representation Engineering to propose a jailbreaking method that does not require elaborate construction prompts, is not affected by model fine-tuning, and can be widely applied to any open-source LLMs in a pluggable manner. We have evaluated this method on multiple mainstream LLMs on carefully supplemented toxicity datasets, and the experimental results demonstrate the significant effectiveness of our approach. After being surprised by some interesting jailbreaking cases, we did extensive in-depth research to explore the techniques behind this method.

Large Language Models (LLMs) have demonstrated great potential as generalist assistants, showcasing powerful task understanding and problem-solving capabilities. To deploy LLMs as AI assistants, it is crucial that these models exhibit desirable behavioral traits, such as non-toxicity and resilience against jailbreak attempts. Current methods for detoxification or preventing jailbreaking usually involve Supervised Fine-Tuning (SFT) or Reinforcement Learning from Human Feedback (RLHF), which requires finetuning billions of parameters through gradient descent with substantial computation cost. Furthermore, models modified through SFT and RLHF may deviate from the pretrained models, potentially leading to a degradation in foundational LLM capabilities. In this paper, we observe that surprisingly, directly editing a small subset of parameters can effectively modulate specific behaviors of LLMs, such as detoxification and resistance to jailbreaking. Specifically, for a behavior that we aim to avoid, we employ a linear classifier, which we term the behavior probe, to classify binary behavior labels within the hidden state space of the LLM. Using this probe, we introduce an algorithm to identify a critical subset of LLM parameters that significantly influence this targeted behavior. Then we directly edit these selected parameters by shifting them towards the behavior probe. Such a direct parameter editing method necessitates only inference-level computational resources. Experiments demonstrate that in the representative detoxification task, our approach achieves reductions of up to 90.0\% in toxicity on the RealToxicityPrompts dataset and 49.2\% on ToxiGen, while maintaining the LLM's general capabilities in areas such as common sense, question answering, and mathematics. Our code is available at https://github.com/lucywang720/model-surgery.

Recent advancements in large language models (LLMs) have led to significant successes across various applications, where the most noticeable is to a series of emerging capabilities, particularly in the areas of In-Context Learning (ICL) and Chain-of-Thought (CoT). To better understand and control model performance, many studies have begun investigating the underlying causes of these phenomena and their impact on task outcomes. However, existing explanatory frameworks predominantly focus on isolating and explaining ICL and CoT independently, leading to an incomplete understanding of their combined influence on model performance. To address this gap, we propose the Electronic Circuit Model (ECM), which provides a foundation for developing scalable, learnable policies and improving the management of AI-generated content. Specifically, ECM conceptualizes model behavior as an electronic circuit: ICL is represented as semantic magnetic field to providing an additional voltage following Faraday's Law, while CoT is modeled as series resistors to constrain the model output performance following Ohm's Law. Experimental results demonstrate that the ECM effectively predicts and explains LLM performance across a variety of prompting strategies. Furthermore, we apply ECM to advanced reasoning strategy optimization on a series of tasks, such as the International Olympiad in Informatics (IOI) and the International Mathematical Olympiad (IMO), achieving competitive performance that surpasses nearly 80% of top human competitors.

Previous works on Large Language Models (LLMs) have mainly focused on evaluating their helpfulness or harmlessness. However, honesty, another crucial alignment criterion, has received relatively less attention. Dishonest behaviors in LLMs, such as spreading misinformation and defrauding users, eroding user trust, and causing real-world harm, present severe risks that intensify as these models approach superintelligence levels. Enhancing honesty in LLMs addresses critical deficiencies and helps uncover latent capabilities that are not readily expressed. This underscores the urgent need for reliable methods and benchmarks to effectively ensure and evaluate the honesty of LLMs. In this paper, we introduce BeHonest, a pioneering benchmark specifically designed to assess honesty in LLMs comprehensively. BeHonest evaluates three essential aspects of honesty: awareness of knowledge boundaries, avoidance of deceit, and consistency in responses. Building on this foundation, we designed 10 scenarios to evaluate and analyze 9 popular LLMs on the market, including both closed-source and open-source models from different model families with varied model sizes. Our findings indicate that there is still significant room for improvement in the honesty of LLMs. We also encourage the AI community to prioritize honesty alignment in LLMs. Our benchmark and code can be found at: https://github.com/GAIR-NLP/BeHonest.