Fired OpenAI safety researchers tell board: don't build models with hard-to-monitor reasoning
Hesamation · x · 2026-10-08
Three safety researchers fired by OpenAI wrote a letter to the board urging the company not to build models with hard-to-monitor reasoning. The letter reveals more firing details: OpenAI claims they were let go for sharing sensitive information externally, but Korbak was OpenAI's contact for METR's Hugging Face hack investigation and Balesni was helping the board build an industry pledge on monitoring AI reasoning — jobs that required constant external communication. This came weeks after Altman voiced support for outside evaluators. They say they never overstepped and that the firings are "chilling everyone still at OpenAI." OpenAI denies the firings related to safety concerns.
Related event: Fired OpenAI Safety Researchers Write to Board, Warning of Chilling Effect(4 posts)→
More from Companies & People
- Researchers from frontier labs debate continual learning for 3 hours — ysu_nlp · 2026-10-08
- Stop rebuilding the agent stack: the case for shipping Claude-native apps — matt_slotnick · 2026-10-08
- India's IICT and YouTube premiere first four shorts from Create with AI programme — lmoroney · 2026-10-08
- Ex-Salesforce AI lead Clara Shih: AI outperforms new grads, so we changed hiring — clarashih · 2026-10-08
- Sequence Holdings CEO: AI's real opportunity is rethinking how companies work — No Priors · 2026-10-08
- HeyGen Makes Paraform's Talent Density Index Top 50, Says It Never Hires to Fill Seats — HeyGen · 2026-10-08