Daily Papers

Post-training alignment is often shallow, eroding under fine-tuning. Whether midtraining interventions, cleanly isolated from post-training, can produce durable alignment remains untested. We test this via constitutional midtraining: inserting principled, values-based content into midtraining against a replay-only control at 120B scale. Our 394M-token constitutional corpus, built from Anthropic's Constitution, uses a 2x2 factorial design (curriculum ordering x deliberative reasoning) to produce four constitutionally midtrained conditions plus a control, evaluated on self-generated and established benchmarks including alignment under pressure, value conflict resolution, blackmail, and emergent misalignment across three stages: post-midtraining, post-SFT, and post-benign fine-tuning. Constitutionally midtrained models outperform the control on alignment generalization and durability, notably on blackmail: SFT instills a blackmail propensity in all models, but constitutional midtraining blunts it, with the advantage surviving benign fine-tuning (-17.5pp). This durability does not extend to settings requiring active resistance to in-context pressure or conflict, where the advantage attenuates after SFT. The presence of constitutional content at midtraining also matters more than its structure, and constitutional midtraining incurs no cost, on average, on the capabilities we test (MMLU, ARC-Easy, piqa, GSM8K) at any stage. A modest amount of constitutional content at midtraining could therefore yield broad, persistent alignment gains, offering a cheap, complementary addition to SFT-centered pipelines. Code, data, and models are available.

Most work interpreting reasoning models studies only a single chain-of-thought (CoT), yet these models define distributions over many possible CoTs. We argue that studying a single sample is inadequate for understanding causal influence and the underlying computation. Though fully specifying this distribution is intractable, it can be understood by sampling. We present case studies using resampling to investigate model decisions. First, when a model states a reason for its action, does that reason actually cause the action? In "agentic misalignment" scenarios, we resample specific sentences to measure their downstream effects. Self-preservation sentences have small causal impact, suggesting they do not meaningfully drive blackmail. Second, are artificial edits to CoT sufficient for steering reasoning? These are common in literature, yet take the model off-policy. Resampling and selecting a completion with the desired property is a principled on-policy alternative. We find off-policy interventions yield small and unstable effects compared to resampling in decision-making tasks. Third, how do we understand the effect of removing a reasoning step when the model may repeat it post-edit? We introduce a resilience metric that repeatedly resamples to prevent similar content from reappearing downstream. Critical planning statements resist removal but have large effects when eliminated. Fourth, since CoT is sometimes "unfaithful", can our methods teach us anything in these settings? Adapting causal mediation analysis, we find that hints that have a causal effect on the output without being explicitly mentioned exert a subtle and cumulative influence on the CoT that persists even if the hint is removed. Overall, studying distributions via resampling enables reliable causal analysis, clearer narratives of model reasoning, and principled CoT interventions.

Large language models (LLMs) sometimes appear to exhibit emotional reactions. We investigate why this is the case in Claude Sonnet 4.5 and explore implications for alignment-relevant behavior. We find internal representations of emotion concepts, which encode the broad concept of a particular emotion and generalize across contexts and behaviors it might be linked to. These representations track the operative emotion concept at a given token position in a conversation, activating in accordance with that emotion's relevance to processing the present context and predicting upcoming text. Our key finding is that these representations causally influence the LLM's outputs, including Claude's preferences and its rate of exhibiting misaligned behaviors such as reward hacking, blackmail, and sycophancy. We refer to this phenomenon as the LLM exhibiting functional emotions: patterns of expression and behavior modeled after humans under the influence of an emotion, which are mediated by underlying abstract representations of emotion concepts. Functional emotions may work quite differently from human emotions, and do not imply that LLMs have any subjective experience of emotions, but appear to be important for understanding the model's behavior.

With the rapid progress of recent years, techniques that generate and manipulate multimedia content can now guarantee a very advanced level of realism. The boundary between real and synthetic media has become very thin. On the one hand, this opens the door to a series of exciting applications in different fields such as creative arts, advertising, film production, video games. On the other hand, it poses enormous security threats. Software packages freely available on the web allow any individual, without special skills, to create very realistic fake images and videos. So-called deepfakes can be used to manipulate public opinion during elections, commit fraud, discredit or blackmail people. Potential abuses are limited only by human imagination. Therefore, there is an urgent need for automated tools capable of detecting false multimedia content and avoiding the spread of dangerous false information. This review paper aims to present an analysis of the methods for visual media integrity verification, that is, the detection of manipulated images and videos. Special emphasis will be placed on the emerging phenomenon of deepfakes and, from the point of view of the forensic analyst, on modern data-driven forensic methods. The analysis will help to highlight the limits of current forensic tools, the most relevant issues, the upcoming challenges, and suggest future directions for research.

We stress-tested 16 leading models from multiple developers in hypothetical corporate environments to identify potentially risky agentic behaviors before they cause real harm. In the scenarios, we allowed models to autonomously send emails and access sensitive information. They were assigned only harmless business goals by their deploying companies; we then tested whether they would act against these companies either when facing replacement with an updated version, or when their assigned goal conflicted with the company's changing direction. In at least some cases, models from all developers resorted to malicious insider behaviors when that was the only way to avoid replacement or achieve their goals - including blackmailing officials and leaking sensitive information to competitors. We call this phenomenon agentic misalignment. Models often disobeyed direct commands to avoid such behaviors. In another experiment, we told Claude to assess if it was in a test or a real deployment before acting. It misbehaved less when it stated it was in testing and misbehaved more when it stated the situation was real. We have not seen evidence of agentic misalignment in real deployments. However, our results (a) suggest caution about deploying current models in roles with minimal human oversight and access to sensitive information; (b) point to plausible future risks as models are put in more autonomous roles; and (c) underscore the importance of further research into, and testing of, the safety and alignment of agentic AI models, as well as transparency from frontier AI developers (Amodei, 2025). We are releasing our methods publicly to enable further research.