arXiv 2502.14143
Multi-Agent Risks from Advanced AI
By Lewis Hammond, Alan Chan, et al.
Published 2025-02-19
Wiki summary
Explore the paper's summary, context, and related research on Papiers.
The rapid development of advanced AI agents and the imminent deployment of many instances of these agents will give rise to multi-agent systems of unprecedented complexity. These systems pose novel and under-explored risks. In this report, we provide a structured taxonomy of these risks by identifying three key failure modes (miscoordination, conflict, and collusion) based on agents' incentives, as well as seven key…