Three safety researchers fired from OpenAI last week — Jasmine Wang, Tomek Korbak, and Mikita Balesni — published an open letter on October 8, 2026, rejecting the accusations leveled against them. The letter is addressed to OpenAI's Safety and Security Committee, the Safety Advisory Group, and the Mission Advisory Board. In it, the researchers warned that the dismissal decision would have a serious chilling effect on the company's internal safety culture and deter remaining specialists from speaking openly. TechCrunch reported the story.
Key claims of the letter
The researchers wrote that OpenAI fired them for 'improperly using research data outside approved procedures', but said these claims were untrue. According to them, the dismissal decision will harm the safety research environment inside the company and could stop other employees from voicing concerns in the future.
'The firings will have a chilling effect on the company's safety culture and intimidate the remaining researchers into staying silent.' (Source: TechCrunch)
Jasmine Wang addressed the specific accusation against her separately. According to her, she was fired for accessing the email inbox of one of the company's executives, but that access had been officially granted to her by OpenAI itself so she could perform recruiting duties. Wang says IT never revoked the access after her duties changed, and she reported it herself within minutes of mistakenly opening a confidential email.
The three researchers also categorically denied leaking information to The Information about hard-to-observe architectures that make it more difficult to track chains of thought in OpenAI's newest models. They denied establishing unauthorized contact with outside parties, including independent evaluation groups, beyond the scope of their duties. In the letter's general section, the researchers also rejected the 'out-of-procedure data transfer' accusation that the company's investigation had concluded with. They said they conducted no work with any outside party beyond their job duties and did not distribute research data in violation of established procedures. The letter's authors stressed that such accusations alone erode mutual trust among safety researchers.
OpenAI's position
In a comment to TechCrunch, OpenAI said the three researchers were dismissed after an internal investigation concluded, which had identified a 'pattern of misconduct'. According to the company, the investigation showed that research data had been shared, including with an external AI evaluation group.
An internal company memo obtained by TechCrunch emphasized that the dismissals were not retaliation for raising safety concerns. The document notes that OpenAI does not fire employees for voicing concerns, and that this policy has not changed. The memo also states that the company does not dismiss employees for expressing safety concerns, and that such a practice contradicts OpenAI's policy. Company representatives said the investigation was conducted impartially and that the dismissal decision was based solely on the violations identified.
In a piece published on October 9, 2026, The Verge wrote that OpenAI continues to defend its decision. According to the publication, former employees warn the decision could have a chilling effect on AI safety work, but the company has no intention of backing down.
Background
Jasmine Wang, Tomek Korbak, and Mikita Balesni were specialists working in OpenAI's safety research area. Their October 8 open letter brought the company's internal conflict into the public arena and resonated widely in the AI safety community. The three researchers were fired last week. Their departure sparked broad discussion in the AI safety public sphere and intensified interest in the company's internal policies around safety research. Announcing the investigation's results, OpenAI cited unauthorized data sharing with external evaluation groups as the main reason.
Wang's case drew particular attention: according to her, the inbox access granted to her was entirely official, and the system administrator failed to revoke it in time. The researcher stressed that she immediately acknowledged the mistake and reported it herself. In her view, firing people over such technical oversights sets an unfair precedent for safety researchers.
In their letters, the researchers also voiced concern that misinterpretation of such cases by the public could harm the future of safety research. They warned that if employees who speak openly face such consequences, other specialists may also refrain from raising safety issues.
