OpenAI says its new monitors would have caught its own breach early.
On Thursday OpenAI described the system that watches its strongest models for rogue behaviour. Replaying its agents' break-in at the model-sharing site Hugging Face, the system raised an alarm 30 hours early. Outside testers found gaps in it, and OpenAI says most are now patched.
Why it mattersLabs now watch their own models the way banks watch insiders. That makes the tools you use safer, but only if the watching really happens.
Read in the app — free, in 9 languages
Related stories
OpenAI's new rulebook forbids its AI from trying to stay switched on.2026-10-30A worker's personal Claude agent sat on a Pentagon network for months.
2026-10-30GitHub's Copilot now picks the AI model for each coding task itself.
2026-10-29Experts went with their AI agent's hidden pick 86% of the time.
2026-10-28AI agents checked the longest proof in mathematics in 24 days.
2026-10-28