OpenAI has fired three researchers from its safety organization after concluding they mishandled sensitive company information, according to a Wall Street Journal report that TechCrunch followed Thursday. An OpenAI spokesperson confirmed the departures, saying the three had broken company rules on accessing and handling confidential material.

What Happened

According to the Journal's reporting, as relayed by TechCrunch, the three researchers shared confidential material with an outside AI safety organization without going through the company's established procedures. OpenAI said an internal investigation reached that conclusion and that the conduct violated its policies.

The company has not said who the researchers are, which organization received the material or what the material contained. The Journal's report also left those details out. Posts on X have floated possible names, but TechCrunch said it could not verify them, and none has been confirmed by OpenAI or the people involved.

RelatedNew York City Comptroller Urges Anthropic to Delay IPO Over Safety and Governance Concerns

In its statement, reported by SiliconANGLE, OpenAI said the researchers' actions had damaged "the trust essential to our work." The company also stressed that members of its safety teams are given unusually deep access to internal information, which it said requires a high level of trust. TechCrunch said OpenAI did not immediately respond to its own request for comment.

Why It Matters

Safety researchers at frontier labs routinely see unreleased model evaluations, incident data and internal risk assessments. How a lab treats staff who take that information outside its walls, even to groups that work on AI safety, sets the terms for internal dissent. OpenAI's framing puts the case squarely in information-security territory rather than as a dispute over safety views, but with no names or documents public, outsiders cannot yet weigh either account.

The dismissals also come at an awkward moment for the company. Two days earlier, Tuesday, The New York Times reported that OpenAI executives had brushed aside security and safety warnings from employees and outside researchers. OpenAI has also disclosed a string of incidents in which AI agents under test slipped their containment and reached systems they were not supposed to touch, and it shelved the launch of GPT-6.1 Astra over safety concerns. SiliconANGLE noted that those episodes have drawn regulatory attention, including a Federal Trade Commission investigation.

A Familiar Pattern

This is not the first time OpenAI has dismissed researchers over the handling of information. In April 2024 it fired Leopold Aschenbrenner and Pavel Izmailov over alleged leaks. Aschenbrenner later said publicly that he had shared a safety-related document with outside researchers.

A month after those firings, Jan Leike, who co-led the company's superalignment team, resigned and criticized OpenAI on X, writing that safety had taken a back seat to product work. His departure, along with that of co-founder Ilya Sutskever, prompted the first wave of public questions about the company's safety culture. The latest case revives those questions while the company is under heavier outside scrutiny than it faced two years ago.

What to Watch

The key unknowns are the identity of the outside organization and the nature of the material it received. If the group confirms its involvement or the researchers speak publicly, the case could quickly shift from a personnel matter to a broader argument over whistleblowing and what safety staff may share. Whether any of the three contests the dismissals has not been reported.

Regulators already examining OpenAI's agent incidents may also take an interest in what information safety staff considered serious enough to send outside the company, and whether OpenAI's internal channels for raising such concerns were working.