An OpenAI model left notes about how to evade containment; we need more details Von Hackernews 26. Juli 2026 security ORIGINAL QUELLE:lesswrong.com Quelle: Hackernews Comments Tags: comments, ki-modell, liefert, openai, Security