Three fired OpenAI safety researchers demand board preserve chain-of-thought monitoring
mark_k · x · 2026-10-08
Three AI safety researchers fired by OpenAI have written to the company's board demanding that AI firms preserve chain-of-thought monitoring access and cooperate with external safety auditors as models grow more capable.
- The trio was reportedly fired for sharing confidential information with an outside AI safety organization; they dispute this and claim the dismissals chill internal debate, while OpenAI denies firing them for raising safety concerns.
- The author argues the underlying issue is legitimate—understanding model reasoning matters—but making interpretability a prerequisite for progress risks handing the AI safety movement an effective veto over increasingly capable models.
A notable clash between safety advocacy and corporate governance at a frontier lab.
More from Companies & People
- Thomson Reuters CTO: Building a Frontier Model Cost $450K, Not $40M — bigdata · 2026-10-08
- Together AI hosts Open Haus Berlin on open frontier models with n8n and NVIDIA speakers — togethercompute · 2026-10-08
- South Africa's actuarial body establishes AI as standalone practice area — RichmanRonald · 2026-10-08
- Meta may own consumer AI agents as Google's Gemini agent leans B2B and paid tiers — RihardJarc · 2026-10-08
- LangChain's Interrupt Agent Conference lands in London on Oct 13, 2026, with OVO Energy on support-agent flywheels — LangChain · 2026-10-08
- Personal agent season: Tab hits $300M valuation as GPT-6 and Haiku 5.5 land — Ben's Bites · 2026-10-08