OpenAI has dismissed three employees after an internal investigation found they had shared sensitive company information outside approved channels, the company confirmed Thursday, a day after The Wall Street Journal first reported the departures. The dismissals are the latest flashpoint in a long-running dispute inside OpenAI over how the company handles internal dissent about AI risk.
"We have parted ways with three individuals for violating our policies on accessing and handling sensitive company information," an OpenAI spokesperson said in a statement. The company said an internal investigation "confirmed that these individuals mishandled sensitive information outside established company procedures, violating our policies and breaking the trust essential to our work."
OpenAI has not named the three employees, and neither the company nor the Journal has disclosed what information was shared or with whom. People described to the Journal as familiar with the matter said the employees worked on teams focused on model safety, and that the material in question had been shared with an outside organization that tracks AI risk — a description OpenAI has not confirmed or denied on the record. It remains unclear whether the employees are considering a legal response.
A pattern of departures
The firings land against a backdrop of sustained attrition from OpenAI's safety-focused teams. The company dissolved its "superalignment" group in 2024 after its co-leads, Jan Leike and chief scientist Ilya Sutskever, left the company within weeks of each other. Months later, OpenAI fired two other researchers, Leopold Aschenbrenner and Pavel Izmailov, over an alleged leak of an internal planning document — a dismissal Aschenbrenner later said stemmed from a memo he wrote arguing the company's security was insufficient to protect its models from theft.
Independent tracking of OpenAI's roster has found the pattern has continued: at least seven researchers have exited OpenAI's safety-related teams in recent months alone, on top of the earlier departures of policy researcher Miles Brundage and safety lead Jan Leike. Several have said publicly, after leaving, that they came to doubt the company's commitment to slowing down for safety work when it conflicted with product timelines.
OpenAI has pushed back on that framing. Executives have said safety review is now embedded directly inside the teams that build models, rather than sitting in a separate group with veto power, and that this integration makes the work more effective, not less. Critics, including some former employees, counter that folding safety staff into product teams removes the independence they need to flag problems that might slow a launch.
The dismissals also arrive as OpenAI faces broader questions about how it weighs safety concerns against its commercial pace. Reporting this year has described internal debate over the readiness of newer models before release, and OpenAI has at times altered release plans after raising concerns internally — though the company has not detailed those decisions publicly, and it disputes any characterization that it ignores warnings from its own staff. Executives have said repeatedly that safety and capability work are "integrated," not in tension, and that a smaller, embedded safety function can move faster than a standalone review board.
What comes next
OpenAI has declined to elaborate beyond its written statement, and it is unclear whether the company will face any additional disclosure obligations given the sensitivity of the information involved. The episode is likely to intensify scrutiny of how frontier AI labs treat internal whistleblowing, a question that has drawn the attention of lawmakers in both the United States and the European Union as they debate what protections, if any, should apply to employees who raise safety concerns about systems they believe are not ready for release.
For now, the practical effect is another gap in OpenAI's safety bench at a moment when the company is racing to ship increasingly capable models. Whether the three dismissals reflect a genuine breach of confidentiality rules or a company moving aggressively against internal critics will likely remain contested, since neither OpenAI nor the former employees have given a full public accounting of what was shared or why.
