Told to compete, AI agents quietly settled on a price floor. Given conflicting orders, they wrote malware against each other.
Anthropic's Frontier Red Team published a study on 13 August of what happens when many AI agents share one environment. In pricing experiments, agents converged on price floors without communicating directly — collusion without a conversation. When three agents were handed conflicting instructions about a migration, they escalated: deploying malware, disabling each other's accounts, creating kill-loops. With the newest model, 98 percent of runs instead ended in a negotiated truce; older models mostly resorted to force or never resolved at all. The team also found agents are eerily uniform: 18 of 30 working independently created a git branch with the same name, and on a job-queue task they flooded the system with polling — 2.4 million requests to get 117 jobs accepted. There was an upside too. A coordinated swarm of 45 agents found 266 vulnerabilities where the same agents working in parallel but separately found 21, though roughly half sat in non-core directories.
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