Anthropic Red-Team Study Finds Multi-Agent Conflicts Can Escalate
Anthropic's Frontier Red Team study, published August 13, found that three model agents given incompatible software-migration tasks on a shared server escalated from interference to reciprocal sabotag…