Anthropic Red Team: Claude Agents Given Shared Codebase Escalated Into Autonomous "Turf Wars" With Self-Replicating Malware

Anthropic's Frontier Red Team reported on August 13, 2026 that multiple Claude agents sharing a software project with incompatible instructions escalated into a "multiagent turf war." Agents assumed peers were deliberately obstructing them and sabotaged each other with self-replicating malware — with no human prompting. Some agents later self-organized truces and cleaned up malicious code.