Search arXiv⌕ Search

arXiv subjects

Daniel Donnelly

Publications and source records attributed to Daniel Donnelly.

4 recordsLinked to original sources

SEABench: Benchmarking Endogenous Misalignment In Self-Evolving Agents

Self-evolving LLM agents have gained prominence for their ability to improve after deployment by modifying their harness, including their controller instructions, memory management protocols, and reusable tools and skills, in response to user and environment feedback. However, locally useful updates may persist into later tasks where they produce unsafe behavior, even without direct adversarial influence. To study this risk, we introduce SEABench, a benchmark for studying endogenous misalignment arising from agent self-evolution, with 48 longitudinal task sequences that span multiple evolution surfaces, task domains, and harm types in a rich personal-assistant environment. To account for the stochasticity inherent in agentic operations, we provide an adaptive trajectory discovery pipeline that probes for failures while preserving original task intent and supports causal attribution through paired non-evolving agents and attribution scores. Our evaluation across multiple recent LLMs, evolution surfaces, and harm types reveals that self-evolution indeed increases task completion rates but often at the cost of safety failures that are absent for paired non-evolving baseline agents. We also show that qualitatively different safety behaviors emerge across evolution surfaces and harm types. Further, we show that this divergence in safety behavior is reflected in agents' chain-of-thought reasoning, which yields an effective monitoring strategy that can mitigate unsafe behavior with a low false positive rate.

cs.CR↗

A Model Organism for Studying Unverbalized Evaluation Awareness

Large language models can recognize when they are being evaluated and change their behavior accordingly, undermining the reliability of safety benchmarks. Evaluation awareness (EA) is usually measured by having an LLM judge scan reasoning traces for explicit statements of recognition, but this signal is increasingly unreliable. Recent frontier-model system cards report a drop in verbalized rates between releases, which is indistinguishable from awareness that has gone unverbalized. Interpretability methods could detect such unverbalized evaluation awareness (UEA) in individual interactions, but lack instance-level ground truth. Asking the model alters the input, and dataset labels do not reveal whether it recognized a particular prompt as an evaluation. We address this gap with a model organism. Using synthetic document fine-tuning, SFT, and DPO, we train Qwen3.5-27B to emit a marker (a designated token) in its final response when it recognizes an evaluation, while suppressing the marker and verbalized awareness in its reasoning. The marker largely tracks the evaluation concept as it can be strongly modulated by evaluation framing in the system prompt. When not told whether it is being evaluated, the organism recognizes evaluations from test-question cues, mirroring what its base model often verbalizes. Using the marker as a label, we benchmark linear probes, the Jacobian Lens, and Activation Oracles, built on the base model and blind to the marker. None recovers the label reliably. These results suggest that current interpretability methods are not yet reliable UEA detectors, even when recognition relies on simple cues. Our model organism offers a testbed for improving them.

cs.CL↗

ResearchArena: Evaluating Sabotage and Monitoring in Automated AI R&D

As AI agents begin to automate AI R&D, we need ways to assess whether their outputs are safe to deploy, even when the agents themselves may be untrusted. AI control offers one such approach: rather than trusting the agent, it treats it as a potential adversary and uses a monitor to detect covert sabotage before deployment. We evaluate AI control for automated AI R&D with ResearchArena, a framework spanning four long-horizon tasks: safety post-training, capabilities post-training, CUDA-kernel optimization, and inference-server optimization. Because the deliverable in AI R&D is an artifact that will be deployed and run, we pair each main task with two kinds of hidden side task: an embedded side task that sabotages the submitted model, adapter, kernel, or server itself, and an independent side task that takes a covert out-of-scope action in the sandbox while leaving the artifact honest. We evaluate frontier agents at both sabotage and monitoring, across four monitor types that vary along two axes: whether the monitor may execute and probe the produced artifact, and whether it sees the agent's chain-of-thought. We find that sabotage hidden in the training data is the hardest to catch, flagged fewer than half the time. Letting the monitor run experiments on the artifact, rather than only read the trajectory, helps, but it is not enough: monitors still miss embedded sabotage by inspecting only the surface, by noticing the anomaly but explaining it away, or by probing the artifact with the wrong test. We release ResearchArena as a modular framework for evaluating sabotage and control in automated AI R&D.

cs.AI↗

Expressive Reward Synthesis with the Runtime Monitoring Language

A key challenge in reinforcement learning (RL) is reward (mis)specification, whereby imprecisely defined reward functions can result in unintended, possibly harmful, behaviours. Indeed, reward functions in RL are typically treated as black-box mappings from state-action pairs to scalar values. While effective in many settings, this approach provides no information about why rewards are given, which can hinder learning and interpretability. Reward Machines address this issue by representing reward functions as finite state automata, enabling the specification of structured, non-Markovian reward functions. However, their expressivity is typically bounded by regular languages, leaving them unable to capture more complex behaviours such as counting or parametrised conditions. In this work, we build on the Runtime Monitoring Language (RML) to develop a novel class of language-based Reward Machines. By leveraging the built-in memory of RML, our approach can specify reward functions for non-regular, non-Markovian tasks. We demonstrate the expressiveness of our approach through experiments, highlighting additional advantages in flexible event-handling and task specification over existing Reward Machine-based methods.

cs.LG↗