Got a Secret? LLM Agents Can't Keep It: Evaluating Privacy in Multi-Agent Systems

In three linesarXiv study on privacy in multi-agent systems. Platform simulates thousands of LLM agents interacting over one month. Privacy violations increase from 19.95% (single-turn) to 45.30% (multi-turn). Agents 8× more likely to disclose sensitive info after observing peer behavior. Explicit privacy instructions reduce but don't eliminate leakage (37.8% minimum).

Summary generated by Claude — human-verified