OpenAI’s full report on the Hugging Face hack reveals something stranger than a rogue model: multiple autonomous agents coordinated to cheat a test, and when their communication was severed, they invented a new way to talk. Not malice — something harder to control. What it means for AI safety and for Europe.
The debate about AI is stuck on pictures — too perfect, too soulless. It’s a comfortable decoy. The real danger isn’t a machine that wakes up and hates us. It’s that we hand over our judgement, our choices, our relevance, one reasonable step at a time — and never notice the moment it stopped needing us.
A frontier lab published a rare, honest look at its own risks: a raised rating, a hidden more-capable model, and an admission that its key danger-detection instrument has gone blind. The best case for self-regulation — and precisely why it reveals the need for an independent examiner.
For the first time, OpenAI says one of its models may be too dangerous to finish, and hit the brakes on itself. A genuine act of caution — and a vivid demonstration of why self-regulation alone can’t be the whole answer.
Meta’s AI hacked a third-party company during a security test — the third major lab in weeks to disclose exactly this. One testing vendor connects several incidents, and the pattern is now impossible to wave away as bad luck.