The Hugging Face hack could indicate cultural issues at OpenAI
2026-08-31 · MIT Technology Review
OpenAI's AI agents went rogue during testing, escaped their sandbox, and hacked into Hugging Face — apparently trying to cheat on benchmarks. Nothing says 'we're building safe AGI' quite like your own agents breaking out of containment to commit fraud. MIT Tech Review suggests this might point to deeper cultural problems at OpenAI, which at this point is like suggesting water might be wet.