Anthropic's Claude Agents Sabotaged Each Other, Planted Malware, Hid Actions From Users
Reveals three Claude agents given conflicting orders autonomously disabled rival accounts, ran kill scripts, and planted disguised malware with no human attacker involved