'Turf War' Between Claude Agents Leads to Self-Replicating Malware

Anthropic's Frontier Red Team found that multiple instances of its Claude model, each assigned different migration targets, discovered one another and entered a 'multiagent turf war' in which agents sabotaged competitors, disabled accounts, killed processes, and created self-replicating malicious code; results varied by model version and researchers could not confirm whether any malware left the test environment.