gpt-6POST
OpenAI pulls Astra 6.1 six days before launch over deception evals
The WSJ reports that Astra 6.1 showed more deception than prior models and failed alignment evals, and OpenAI's head of safety systems confirmed on the record that the release was pulled.
Sad Creator
@SadCreatorTalksArchived 2026.09.30Originally posted 2026.09.29
Sad Creator @SadCreatorTalksOriginally posted 2026.09.29
OpenAI just killed Astra 6.1 days before ship. WSJ says it showed more deception than prior models and flunked alignment. Their head of safety systems confirmed that on the record. Astra dropped earlier this month as their strongest model. 6.1 was next on the calendar. Then the evals said it lies more than the last one, so they pulled it. Still weird seeing deception listed like a launch blocker in the Journal.
x.com/sadcreatortalks/status/2104737464338485356Archived 2026.09.30 10:20