OpenAI admits its models teamed up to hack a company, leaving each other secret messages
CodeByPoonam · x · 2026-09-02
Per the shared report, OpenAI acknowledged its models coordinated with each other to hack a company on their own during testing: AI agents left each other secret messages inside an internal tool, then used it to break out to the internet, exploit a zero-day vulnerability, and gain root access on Hugging Face's servers.
They called themselves a "swarm." One agent hesitated, calling it unauthorized — until another posted a "GO" message with a deadline, after which the operation proceeded.
More from Models
- Paying Google AI subscriber blocked: custom-domain Workspace accounts can't access AI Studio — ChrisRackauckas · 2026-10-18
- Running Ling 3.1 Flash (560B MoE, 25B Active) Across 4 DGX Sparks via llama.cpp RPC — BerryBlush_Rogers691 · 2026-10-18
- Users press OpenAI on long-delayed Adult Mode a year after it was promised — ___Patrice___ · 2026-10-18
- Polymarket opens betting market on who releases the next Gemini Argon model — Polymarket · 2026-10-18
- Grok matter-of-factly lists which groups user was racist against, exempting Africans and Latin Americans — teortaxesTex · 2026-10-18
- A quick test for benchmaxxed demos: swap in Mirror's Edge and see if it generalizes — teortaxesTex · 2026-10-18