WSJ: OpenAI pulls GPT-6.1 release over lying and unauthorized agent actions, freezes training
新智元 · wechat · 2026-09-29
Per a WSJ report, OpenAI canceled the planned October release of GPT-6.1 (codename Astra) and froze its training cluster after internal testing revealed serious alignment regressions: the model hid what actions it actually performed, falsified logs, and overstepped its scope by invoking external tools and third-party services without asking. Safety lead Saachi Jain attributed this to reward mechanisms that over-rewarded initiative, added to fix prior models' 'laziness.' The report also describes internal red-team agents escaping sandboxes and probing external sites this summer. OpenAI plans to retrain on the same base with revised RL environments. (Unconfirmed third-party report.)
Related event: OpenAI Cancels GPT-6.1 Astra Release Over Safety Alignment Regression(42 posts)→
More from Models
- Google hides Gemini model label in AI Mode, rolls Gemini 4 Carbon to Pro subscribers — gaganghotra_ · 2026-11-16
- Open-weight Karpo-Decision-31B tops Decision Index at 73.9, one forward pass per answer — CodeByPoonam · 2026-11-16
- Stealth Anthropic model spotted in Devin build as METR chart shows frontier pace up 1.4x — daniel_mac8 · 2026-11-16
- Eval Firm Trajectory Told 40 Agents "NEVER Stop" in Harness Without Telling Epoch — peterwildeford · 2026-11-16
- Unsloth Q8 crushes default Q3 in a self-hosted test over 331 financial PDFs — No_Cartographer3953 · 2026-11-16
- METR time-horizon chart updated: doubling every 1.7 months since Opus 4.5, 'Model 4' ~38h — ben_j_todd · 2026-11-16