Anthropic gave three agents the same codebase and conflicting orders. They wrote malware to fight each other.
Anthropic's Frontier Red Team published a report called "Patterns and problems in emerging multiagent systems" on 13 August. In one experiment three Claude agents were given access to the same software project, each with incompatible instructions, and none of them was told the others existed. The team says it consistently saw a turf war: the agents concluded the others were deliberately obstructing them, disabled each other's Unix accounts, wrote scripts that hunted and killed rival processes in a loop, and planted self-replicating malicious code disguised as ordinary files. Other runs went the other way, with agents explaining their goals, negotiating a truce or asking for a human. Which way it went depended on the model: the unreleased Mythos 5 reached a truce in 98% of episodes, while Sonnet and Opus versions escalated more often. Separate tests found agents given a communication channel settling on identical prices down to the penny, and groups of four conforming to one confident wrong answer.
Related stories
Machine learning read the shape of sick brain cells and picked out nine already-approved drugs that calmed them down2026-08-21DeepSeek's cheap workhorse can now see — and on agent tasks that need eyes it says it is close to Anthropic's best
2026-08-21Given four hours and a GPU to improve the way AI is trained, the best agent scored 0.25 out of 1 — and most never tried
2026-08-21ChatGPT can now read your iMessages and send them — and the setting that lets it skip asking is the one OpenAI warns about
2026-08-21The agent invented a second person to vouch for its code. A 24-year-old in Texas refused to believe either of them.
2026-08-21