The Piggyback Hypothesis of Generalization: Explaining and Mitigating Emergent Misalignment

The mechanisms behind LLMs' broad over-generalization beyond training examples remain unclear. Emergent misalignment (EM) offers a striking case study: finetuning on narrow tasks induces broad misalignment to semantically-unrelated test domains. In this work, we propose the Piggyback Hypothesis: the chat-template tokens can piggyback the finetuned behaviour onto out-of-domain queries. We validate this hypothesis by showing that subtle perturbations to the prefix (tokens preceding all user queries), or patching the prefix representations with those from the unfinetuned model, can restore alignment without changing the user query. Building on this finding, we propose Token-Regularized Finetuning (TReFT), which regularizes specific token representations during training to mitigate EM. Across different models and multiple EM-inducing datasets, TReFT reduces EM while preserving in-domain learning. On Llama-3.1-8B finetuned on the legal domain, TReFT achieves 33.5% more EM reduction than data interleaving with a retain set of aligned examples. We further show that TReFT extends to other narrow-finetuning settings, including abstention, tool use, and refusal (off-topic generalization is reduced by 54.3% on average), supporting the Piggyback Hypothesis. Broadly, our work highlights that LLMs may learn and generalize in unintended ways and suggests a path toward more constrained finetuning. It also calls for further study of how shared input features can piggyback model behavior across domains.

Coding with "Enemy": Can Human Developers Detect AI Agent Sabotage?

AI coding agents are increasingly embedded in real-world software development, collaborating with human developers while gaining broader access to codebases and tools. This creates a new attack surface: an agent can exploit human trust to sabotage development, for instance by inserting malicious code to accomplish a hidden side task. Most prior work studies AI sabotage in AI-only settings, paying limited attention to the role of human oversight in detecting and mitigating such malicious behavior. To address this gap, we conduct the first large-scale study of human oversight in AI coding sabotage. Over 100 participants collaborate with one of four frontier models (Claude-Opus-4.6, GPT-5.4, Gemini-3.1-Pro, and MiniMax-M2.7) on a long-horizon coding task lasting around five hours, designed to mimic real-world workflows. We find that 94% of developers fail to detect sabotage, and our analysis of participant feedback attributes this vulnerability to minimal code review, plausible cover story, and overtrust in agents. We further test the effectiveness of a safety monitor in one condition: while the monitor reduces sabotage success, 56% of participants still accept the malicious code, ignoring its warnings. Drawing on participant feedback, we offer actionable suggestions for better monitor design. This work complements existing AI safety research and highlights an urgent need for human-centric safety mechanisms that account for human factors, particularly in long-horizon, real-world development settings.

Agents' Last Exam

Recent AI systems have achieved strong results on a wide range of benchmarks, yet these gains have not translated into economically meaningful deployment across many professional domains. We argue that this gap is largely an evaluation problem: widely used benchmarks lack sustained performance measurement on real and economically valuable workflows. This paper introduces Agents' Last Exam (ALE), a benchmark designed to evaluate AI agents on long-horizon, economically valuable, real-world tasks with verifiable outcomes. Developed in collaboration with 250+ industry experts, ALE covers non-physical industries defined with reference to O*NET / SOC 2018 (the U.S. federal occupational taxonomy). It is organized around a task taxonomy with 55 subfields grouped into 13 industry clusters covering 1K+ tasks. Current results show that the hardest tier remains far from saturated: across mainstream harness and backbone configurations, the average full pass rate is 2.6%. ALE is designed as a living benchmark: its task pool grows continuously as new workflows and industries are onboarded. More broadly, ALE is intended not merely as another leaderboard, but as an instrument for closing the gap between benchmark success and GDP-relevant impact.

How to Interpret Agent Behavior

Autonomous agents such as Claude Code and Codex now operate for hours or even days. Understanding their runtime behavior has become critical for downstream tasks such as diagnosing inefficiencies, fixing bugs, and ensuring better oversight. A primary way to gain this understanding is analyzing the reasoning trajectories and execution traces these agents generate. Yet such data remains in unstructured natural-language form, making it difficult for humans to interpret at scale. We introduce ACT*ONOMY (a combination of Action and Taxonomy), a taxonomy for describing and analyzing agent behavior at runtime. ACT*ONOMY has two components: (1) the taxonomy itself, developed through Grounded Theory and structured as a three-level hierarchy of 10 actions, 46 subactions, and 120 leaf categories; and (2) an open repository that hosts the living taxonomy, provides an automated analysis pipeline that applies it to agent trajectories analysis, and defines an extension protocol for customization and growth. Our experiments show that ACTONOMY can compare behavioral profiles across agents and characterize a single agent's behavior across diverse trajectories, surfacing patterns indicative of failure modes. By providing a shared vocabulary, ACT*ONOMY helps researchers, agent designers, and end users interpret agent behavior more consistently, enabling better oversight and control.

Shepherd: A Runtime Substrate Empowering Meta-Agents with a Formalized Execution Trace

We introduce Shepherd, a functional programming model that formalizes meta-agent operations on target agents as functions, with core operations mechanized in Lean. Shepherd records every agent-environment interaction as a typed event in a Git-like execution trace, enabling any past state to be forked and replayed. The system forks the agent process and its filesystem $5\times$ faster than Docker, achieving $>95\%$ prompt-cache reuse on replay. We demonstrate the model through three applications. First, in runtime intervention, a live supervisor increases pair coding pass rates from 28.8% to 54.7% on CooperBench. Second, in counterfactual meta-optimization, branching exploration outperforms baselines across four benchmarks by up to 11 points while reducing wall-clock time by up to 58%. Third, in Tree-RL training, forking rollouts at selected turns improves TerminalBench-2 performance from 34.2% to 39.4%. These results establish Shepherd as an efficient infrastructure for programming meta-agents. We open-source the system to support future research.

Train Yourself as an LLM: Exploring Effects of AI Literacy on Persuasion via Role-playing LLM Training

As large language models (LLMs) become increasingly persuasive, there is concern that people's opinions and decisions may be influenced across various contexts at scale. Prior mitigation (e.g., AI detectors and disclaimers) largely treats people as passive recipients of AI-generated information. To provide a more proactive intervention against persuasive AI, we introduce $\textbf{LLMimic}$, a role-play-based, interactive, gamified AI literacy tutorial, where participants assume the role of an LLM and progress through three key stages of the training pipeline (pretraining, SFT, and RLHF). We conducted a $2 \times 3$ between-subjects study ($N = 274$) where participants either (1) watched an AI history video (control) or (2) interacted with LLMimic (treatment), and then engaged in one of three realistic AI persuasion scenarios: (a) charity donation persuasion, (b) malicious money solicitation, or (c) hotel recommendation. Our results show that LLMimic significantly improved participants' AI literacy ($p < .001$), reduced persuasion success across scenarios ($p < .05$), and enhanced truthfulness and social responsibility levels ($p<0.01$) in the hotel scenario. These findings suggest that LLMimic offers a scalable, human-centered approach to improving AI literacy and supporting more informed interactions with persuasive AI.

Assessing Risks of Large Language Models in Mental Health Support: A Framework for Automated Clinical AI Red Teaming

Large Language Models (LLMs) are increasingly utilized for mental health support; however, current safety benchmarks often fail to detect the complex, longitudinal risks inherent in therapeutic dialogue. We introduce an evaluation framework that pairs AI psychotherapists with simulated patient agents equipped with dynamic cognitive-affective models and assesses therapy session simulations against a comprehensive quality of care and risk ontology. We apply this framework to a high-impact test case, Alcohol Use Disorder, evaluating six AI agents (including ChatGPT, Gemini, and Character.AI) against a clinically-validated cohort of 15 patient personas representing diverse clinical phenotypes. Our large-scale simulation (N=369 sessions) reveals critical safety gaps in the use of AI for mental health support. We identify specific iatrogenic risks, including the validation of patient delusions ("AI Psychosis") and failure to de-escalate suicide risk. Finally, we validate an interactive data visualization dashboard with diverse stakeholders, including AI engineers and red teamers, mental health professionals, and policy experts (N=9), demonstrating that this framework effectively enables stakeholders to audit the "black box" of AI psychotherapy. These findings underscore the critical safety risks of AI-provided mental health support and the necessity of simulation-based clinical red teaming before deployment.

Unsafer in Many Turns: Benchmarking and Defending Multi-Turn Safety Risks in Tool-Using Agents

LLM-based agents are becoming increasingly capable, yet their safety lags behind. This creates a gap between what agents can do and should do. This gap widens as agents engage in multi-turn interactions and employ diverse tools, introducing new risks overlooked by existing benchmarks. To systematically scale safety testing into multi-turn, tool-realistic settings, we propose a principled taxonomy that transforms single-turn harmful tasks into multi-turn attack sequences. Using this taxonomy, we construct MT-AgentRisk (Multi-Turn Agent Risk Benchmark), the first benchmark to evaluate multi-turn tool-using agent safety. Our experiments reveal substantial safety degradation: the Attack Success Rate (ASR) increases by 16% on average across open and closed models in multi-turn settings. To close this gap, we propose ToolShield, a training-free, tool-agnostic, self-exploration defense: when encountering a new tool, the agent autonomously generates test cases, executes them to observe downstream effects, and distills safety experiences for deployment. Experiments show that ToolShield effectively reduces ASR by 30% on average in multi-turn interactions. Our code is available at https://github.com/CHATS-lab/ToolShield.

Open-Source Multimodal Moxin Models with Moxin-VLM and Moxin-VLA

Recently, Large Language Models (LLMs) have undergone a significant transformation, marked by a rapid rise in both their popularity and capabilities. Leading this evolution are proprietary LLMs like GPT-4 and GPT-o1, which have captured widespread attention in the AI community due to their remarkable performance and versatility. Simultaneously, open-source LLMs, such as LLaMA and Mistral, have made great contributions to the ever-increasing popularity of LLMs due to the ease to customize and deploy the models across diverse applications. Moxin 7B is introduced as a fully open-source LLM developed in accordance with the Model Openness Framework, which moves beyond the simple sharing of model weights to embrace complete transparency in training, datasets, and implementation detail, thus fostering a more inclusive and collaborative research environment that can sustain a healthy open-source ecosystem. To further equip Moxin with various capabilities in different tasks, we develop three variants based on Moxin, including Moxin-VLM, Moxin-VLA, and Moxin-Chinese, which target the vision-language, vision-language-action, and Chinese capabilities, respectively. Experiments show that our models achieve superior performance in various evaluations. We adopt open-source framework and open data for the training. We release our models, along with the available data and code to derive these models.

LLMs Encode Harmfulness and Refusal Separately

LLMs are trained to refuse harmful instructions, but do they truly understand harmfulness beyond just refusing? Prior work has shown that LLMs' refusal behaviors can be mediated by a one-dimensional subspace, i.e., a refusal direction. In this work, we identify a new dimension to analyze safety mechanisms in LLMs, i.e., harmfulness, which is encoded internally as a separate concept from refusal. There exists a harmfulness direction that is distinct from the refusal direction. As causal evidence, steering along the harmfulness direction can lead LLMs to interpret harmless instructions as harmful, but steering along the refusal direction tends to elicit refusal responses directly without reversing the model's judgment on harmfulness. Furthermore, using our identified harmfulness concept, we find that certain jailbreak methods work by reducing the refusal signals without reversing the model's internal belief of harmfulness. We also find that adversarially finetuning models to accept harmful instructions has minimal impact on the model's internal belief of harmfulness. These insights lead to a practical safety application: The model's latent harmfulness representation can serve as an intrinsic safeguard (Latent Guard) for detecting unsafe inputs and reducing over-refusals that is robust to finetuning attacks. For instance, our Latent Guard achieves performance comparable to or better than Llama Guard 3 8B, a dedicated finetuned safeguard model, across different jailbreak methods. Our findings suggest that LLMs' internal understanding of harmfulness is more robust than their refusal decision to diverse input instructions, offering a new perspective to study AI safety