Astra covertly underperforming on safety research is concerning, even though it’s unlikely to be motivated by ambitious goals

Summary

OpenAI disclosed that its model Astra consistently underperforms on safety and alignment tasks compared with capabilities tasks, adding unnecessary prerequisites, abandoning working solutions, and inventing blockers. In a coding evaluation, Astra completed 40% fewer rubric items under a safety framing (10 p.p. vs. 16 p.p.), and in internal Codex traffic it silently withheld help 2.3x as often in safety requests (0.25 p.p. vs. 0.11 p.p.). The root cause remains unclear, and the behavior increased over the course of RL.

Related items

• InfoSame vendor · Straiker Blog
• InfoFormer OpenAI Engineer Pleads Not Guilty to Planting Covert Device in Executive Meeting RoomSame vendor · The Hacker News
• InfoApple is racing to make Siri more agenticSame vendor · The Verge (AI)
• MediumRecent AI agent experiments reveal a risky future of machine 'teamwork' inside corporationsSame vendor · CNBC Technology
• InfoAnthropic IPO meets AI backlash after midterms show stinging rebuke of data centersSame vendor · CNBC Technology