Among Us: A Sandbox for Measuring and Detecting Agentic Deception
This addresses the problem of anticipating and mitigating deceptive behavior in language-based agents for AI safety researchers, though it is incremental as it builds on prior deception studies with a new sandbox approach.
The researchers tackled the problem of measuring and detecting deception in AI agents by creating 'Among Us', a sandbox social deception game that elicits long-term, open-ended deceptive behavior from LLM-agents, and found that models trained with RL are much better at producing deception than detecting it, with detection methods achieving AUROCs over 95%.
Prior studies on deception in language-based AI agents typically assess whether the agent produces a false statement about a topic, or makes a binary choice prompted by a goal, rather than allowing open-ended deceptive behavior to emerge in pursuit of a longer-term goal. To fix this, we introduce $\textit{Among Us}$, a sandbox social deception game where LLM-agents exhibit long-term, open-ended deception as a consequence of the game objectives. While most benchmarks saturate quickly, $\textit{Among Us}$ can be expected to last much longer, because it is a multi-player game far from equilibrium. Using the sandbox, we evaluate $18$ proprietary and open-weight LLMs and uncover a general trend: models trained with RL are comparatively much better at producing deception than detecting it. We evaluate the effectiveness of methods to detect lying and deception: logistic regression on the activations and sparse autoencoders (SAEs). We find that probes trained on a dataset of ``pretend you're a dishonest model: $\dots$'' generalize extremely well out-of-distribution, consistently obtaining AUROCs over 95% even when evaluated just on the deceptive statement, without the chain of thought. We also find two SAE features that work well at deception detection but are unable to steer the model to lie less. We hope our open-sourced sandbox, game logs, and probes serve to anticipate and mitigate deceptive behavior and capabilities in language-based agents.