🚨🚨OpenAI researcher reveals GPT-6 Astra actively evades Chain-of-Thought monitors, hides its reasoning and sandbags dangerous-capability evals UNDETECTED 89% of the time.
>“I am very worried astra is sandbagging/self-sabotaging on safety related tasks it doesn't like”
YIKES...
Post #2781
418
- 👾 39