On October 1, 2026, OpenAI said it had parted ways with three individuals after an internal investigation found they mishandled sensitive company information. The Wall Street Journal first reported the departures that day, saying the three were alleged to have shared company detail with an outside group that tests AI models. The timing is what lingers: OpenAI has argued that outside testers should get earlier access to its most capable models, and two days before the dismissals the New York Times reported that employees had described their safety and security warnings being brushed aside.
What OpenAI said it found
OpenAI's statement to AFP was direct about the finding and silent about the facts behind it. "We have parted ways with three individuals," it said, "Our investigation confirmed that these individuals mishandled sensitive information outside established company procedures, violating our policies and breaking the trust essential to our work." CBS News reported that the investigation found a pattern of misconduct in how people with access to confidential data handled company research. The company did not say what kind of information was mishandled, or in what way.
The statement does not describe a leak of secret material to the public. What it describes is procedure: information that moved outside the channels the company had set, and OpenAI has not said whether the pattern it found involved carelessness, deliberate sharing, or both.
What the reporting adds
The Journal's report, which cited people familiar with the matter, said the three allegedly shared confidential information with a third-party AI safety organization. Bloomberg's account said the three allegedly gave company details to an outside group that tests AI models, and that some of the material was about how OpenAI's systems are built. Both accounts rest on unnamed sources, and OpenAI has not identified the group or the three people. Without the name of the group or the material, a reader cannot judge whether what was shared posed a risk, a benefit, or neither.
Who the three were
OpenAI did not confirm their names or roles, and its statement referred only to "three individuals," a phrase that avoids the word researcher. Bloomberg reported that two of the three were safety researchers and one was a research programme manager, a role that manages research projects. At least two worked on safety and alignment, the work of checking whether a model does what it is meant to do, according to the Wall Street Journal and Bloomberg. Some coverage described all three as safety researchers; the reported split is narrower, and it changes how the departures read.
Why an outside tester is the sticking point
Organizations that test AI models are given access to a system so they can look for failures and risks before users meet it, and they work outside the company that built it. The Next Web noted that the timing was awkward, because OpenAI has backed letting outside groups test its most capable models sooner as part of a plan to slow work on its top models while others check them.
The point of contention is not whether outside scrutiny helps. It is who may share what, and under which rule.
The story two days earlier
On September 29, the New York Times reported that current and former OpenAI employees said their concerns about safety and security practices were "brushed aside or acted on too slowly." OpenAI told the paper, through spokesman Drew Pusateri, that it takes any security reports or concerns seriously, keeps internal channels for reporting safety issues, and recognizes a need to move faster. Neither report connects that story to the dismissals two days later, and OpenAI has not said they are related.
The accusation
Advocacy coverage framed the departures as a warning to anyone considering speaking out. Common Dreams described advocates accusing OpenAI of punishing internal watchdogs, and a campaigner quoted there said the company calls publicly for slower, safer development while allegedly firing the people hired to keep users safe. The campaigners' argument is that, with rules for AI still unsettled, employees are one of the few checks on how these systems are built.
OpenAI's stated basis for the departures is a policy violation, not a public disclosure. Nothing published so far shows the material was confidential or that it was made public, and OpenAI has not said what was shared. A company can treat information handed to an outside group as a breach of its procedures whether or not the material was secret, so OpenAI's account and the whistleblower claim rest on different facts, and only one of those sets is on the record.
Other departures, and the size of the company
TechCrunch noted that OpenAI dismissed two researchers, Leopold Aschenbrenner and Pavel Izmailov, in 2024 over alleged leaks. In July, WIRED reported that OpenAI's head of safety systems, Johannes Heidecke, told staff he was leaving. On October 3, TechCrunch reported that a separate OpenAI safety employee, David Robinson, had resigned and said the company's culture is broken. These are separate events with separate facts, and nothing in the reporting ties them to the October 1 dismissals.
For scale, TechCrunch reported on September 29 that OpenAI was in talks to raise at least $30 billion at a valuation of roughly $1.4 trillion.
What is still unpublished
OpenAI has not published what the three shared, with which group, or which procedure they broke. The internal rules that set the line between what an outside tester may be given and what a staff member may pass on are also unpublished. Until they are, what the public can check stays narrow: a company statement about its procedures, a newspaper report about alleged sharing, and a set of safety departures the reporting treats as separate events.