AgentMisalignment: Measuring the Propensity for Misaligned Behaviour in LLM-Based Agents

As Large Language Model (LLM) agents become more widespread, associated misalignment risks increase. Prior work has examined agents' ability to enact misaligned behaviour (misalignment capability) and their compliance with harmful instructions (misuse propensity). However, the likelihood of agents attempting misaligned behaviours in real-world settings (misalignment propensity) remains poorly understood. We introduce a misalignment propensity benchmark, AgentMisalignment, consisting of a suite of realistic scenarios in which LLM agents have the opportunity to display misaligned behaviour. We organise our evaluations into subcategories of misaligned behaviours, including goal-guarding, resisting shutdown, sandbagging, and power-seeking. We report the performance of frontier models on our benchmark, observing higher misalignment on average when evaluating more capable models. Finally, we systematically vary agent personalities through different system prompts. We find that persona characteristics can dramatically and unpredictably influence misalignment tendencies -- occasionally far more than the choice of model itself -- highlighting the importance of careful system prompt engineering for deployed AI agents. Our work highlights the failure of current alignment methods to generalise to LLM agents, and underscores the need for further propensity evaluations as autonomous systems become more prevalent.

End-to-End Imitation Learning for Optimal Asteroid Proximity Operations

Controlling spacecraft near asteroids in deep space comes with many challenges. The delays involved necessitate heavy usage of limited onboard computation resources while fuel efficiency remains a priority to support the long loiter times needed for gathering data. Additionally, the difficulty of state determination due to the lack of traditional reference systems requires a guidance, navigation, and control (GNC) pipeline that ideally is both computationally and fuel-efficient, and that incorporates a robust state determination system. In this paper, we propose an end-to-end algorithm utilizing neural networks to generate near-optimal control commands from raw sensor data, as well as a hybrid model predictive control (MPC) guided imitation learning controller delivering improvements in computational efficiency over a traditional MPC controller.