OpenAI Cleans House

Author: Beatrice Nolan URL: https://fortune.com/2026/10/05/openai-safety-firings-raise-awkward-questions/ One-line: OpenAI fired three safety researchers for allegedly sharing confidential information with a third-party AI safety organization, raising questions about whether the company is structurally capable of independent safety oversight.

Key claims

• OpenAI fired Jasmine Wang, Tomek Korbak, and Mikita Balesni — three safety-focused researchers — for allegedly sharing confidential information with an outside AI safety organization. The firings appear to be connected to an ongoing investigation of frontier AI behavior.
• David Robinson, who led OpenAI’s safety reporting efforts, resigned voluntarily and subsequently wrote a piece in The Atlantic about his concerns with OpenAI’s safety culture and organizational structure. His departure combines with the three firings to create a significant safety team exodus.
• Tomek Korbak was OpenAI’s technical contact with METR (the Machine Intelligence Research Institute spinoff focused on dangerous capability evaluations) during the Hugging Face investigation, which adds significance to his departure — he had direct visibility into external safety assessments.
• Multiple institutional actors are now scrutinizing OpenAI simultaneously: the FTC has an ongoing investigation, California Attorney General Rob Bonta issued a subpoena, and the White House has extracted a voluntary “safety pact” from major AI labs. This represents a significant escalation in regulatory attention compared to 18 months earlier.
• The firings create a structural paradox: if safety researchers cannot share information about safety concerns with independent oversight bodies, the only mechanism for accountability runs entirely through OpenAI’s internal processes — which are the very processes the researchers apparently had concerns about.
• The confidential-information justification is common in tech firings but especially fraught here because safety research often requires external corroboration. Unlike product roadmap leaks, sharing safety-relevant information with another safety organization is arguably in the public interest, and the company’s position is that it is not.

Notable quotes

“OpenAI has fired three researchers from its safety team for sharing confidential information with a third-party AI safety organization. The move raises pointed questions about how independent oversight of OpenAI’s most sensitive work is actually supposed to function.”

“David Robinson, who led safety reporting at OpenAI, has resigned and written about his concerns. When the person whose job was to document safety issues chooses to leave and write publicly, that is a signal worth taking seriously.”

“Tomek Korbak was OpenAI’s technical contact with METR — the external organization conducting dangerous capability evaluations. His departure means that relationship will need to be rebuilt from scratch.”

“The FTC investigation, the California AG subpoena, and the White House safety pact are not independent events. They represent a coordinated escalation of institutional attention on the same set of concerns.”

“There is a notable irony: the researchers fired for sharing safety information are now free to share it publicly, in a way that internal employees are not. OpenAI has potentially accelerated the disclosure it sought to prevent.”

How it connects

• OpenAI — the company at the center of this safety governance story
• AI Safety — the broader field these researchers were contributing to
• METR — the external safety evaluation organization that Korbak worked with; his termination disrupts that relationship
• Sam Altman — OpenAI CEO during this episode
• FTC — has an ongoing investigation into OpenAI
• Rob Bonta — California AG who issued a subpoena to OpenAI
• Eliezer Yudkowsky — the AI safety tradition from which many OpenAI safety researchers originate