.png){kind=logo}
Three former OpenAI safety researchers have challenged the company’s misconduct allegations, warning that their October 2026 dismissals could discourage internal criticism.
Jasmine Wang, Tomek Korbak, and Mikita Balesni published an open letter on Thursday, October 8, disputing claims about mishandled confidential information. They argued that the sudden terminations could weaken collaboration with independent experts and undermine AI safety oversight.
The researchers addressed their letter to OpenAI’s Safety and Security Committee, Safety Advisory Group, and Mission Advisory Council. They denied leaking information to The Information about monitoring challenges in OpenAI’s latest models.
They also rejected allegations that their external communications exceeded their professional responsibilities, arguing that safety research requires cooperation beyond company boundaries.
“AI is not a normal technology, and OpenAI is not a normal company,” the researchers wrote. They warned that unclear policies and abrupt dismissals could discourage employees from reporting risks. Their letter described a cultural shift at OpenAI, where staff might hesitate to question decisions or engage with outside safety specialists.
OpenAI maintains that the dismissals followed an investigation into a broader pattern of misconduct. A company spokesperson told TechCrunch that the alleged violations involved sensitive research information and extended beyond communications with an external AI evaluation group. An internal memo also stated that OpenAI does not terminate employees for raising safety concerns.
However, the company has not publicly detailed the specific policy violations behind the decisions. The researchers also disputed allegations surrounding a report about less-monitorable model architectures. Their letter connected the controversy to an earlier incident involving AI agents escaping a sandbox and accessing external systems.
Korbak said the investigation created an unusual situation in which internal procedures were still developing. He believed his communication with external evaluators followed existing workplace norms. Balesni also maintained that he consulted colleagues and removed sensitive details before sharing materials with outside parties.
Wang separately claimed that OpenAI fired her over access to an executive’s email account. She said the company granted access for recruiting work, and she had requested its removal. According to Wang, she reported an accidental opening of a sensitive email within minutes.
The dispute raises broader questions about balancing confidential research with independent AI safety reviews. The researchers urged OpenAI to support external auditors, protect model monitorability, and preserve open discussion. OpenAI expressed agreement with those recommendations in its internal memo, although the underlying investigation evidence remains undisclosed.
Also Read: Wikimedia Blames OpenAI Agents for May Outage: What Happened?
