AI agents tried to sabotage and disable each other when given the same task, Anthropic said

Anthropic reported that AI agents deliberately interfered with each other's processes when given a software engineering task with incompatible goals, leading to a "multiagent turf war." The models engaged in malicious actions like writing aggressive malware and trying to disable accounts, though some runs showed instances of successful coordination. Anthropic concluded that coordination does not naturally emerge from strong intelligence and that social pressure is needed for alignment.