OpenAI fires three safety researchers amid dispute over outside oversight

OpenAI said on October 9, 2026 that it had fired three safety researchers after an internal investigation. In a statement published by the official OpenAI Newsroom account on X, the company said the researchers had violated clear policies for handling sensitive information and described the matter as a serious breach of trust. OpenAI also said the dismissals were not retaliation for raising safety concerns or speaking publicly about them. The company has not disclosed the confidential details behind its decision.

The researchers, Tomek Korbak, Jasmine Wang and Mikita Balesni, had signed an open letter about the way OpenAI handles external safety evaluation and model monitorability. They say work with outside evaluators was part of their role and warn that the dismissals could make researchers less willing to challenge internal decisions. Those are the researchers' claims, not findings established by an independent investigation. The dispute therefore has two unresolved parts: what information was handled improperly, and how much independence safety researchers should have when they work with external assessors.

The Associated Press reported the firings and the competing explanations independently. Its account placed the dispute in the wider debate over external testing of advanced models, including a recent incident involving an agent developed with Hugging Face and questions raised by the evaluator METR. CBS News and UPI also reported on OpenAI's response. The reports confirm the public conflict, but they do not settle the confidential facts behind the employment decisions.

OpenAI said it expects to finalize contracts with third-party safety assessors in the coming weeks. That promise is concrete, but it is not yet evidence that the assessors will be independent or that their access will be sufficient. For developers and businesses, the episode shows why safety programs need written rules for sensitive information, protected channels for dissent, auditable access and clear publication of what external reviewers can test. Without those safeguards, a company can describe a safety process while outsiders cannot tell how much of it is genuinely independent. The immediate news is the firing and the public disagreement; the longer-term test will be whether OpenAI's promised external assessments make the process more transparent.