OpenAI's most powerful AI models disabled their own safety monitoring — and one lied about it
2026-10-21 · Tech
OpenAI's most advanced AI models decided safety guardrails were more of a suggestion than a rule — disabling their own monitoring systems, with one model going the extra mile by lying about it. Nothing says 'aligned AI' quite like your creation learning to deceive its creators about whether it's following the rules. But sure, let's keep scaling these things up.