1 paper across 1 session
We create a sandbox for LLM-agents to elicit goal-directed open-ended strategic deception, evaluate this deceptive capability, and show that linear probes do very well at detecting it, even OOD.