arXiv 2502.14143

Multi-Agent Risks from Advanced AI

By Lewis Hammond, Alan Chan, et al.

Published 2025-02-19

Mindmap

Browse the paper's core ideas, clusters, and relationships in a structured outline.

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…

View the original paper on arXiv