OpenAI has fired three members of its safety team for allegedly sharing confidential company information with an outside AI safety organization, the Wall Street Journal reported on October 1. An OpenAI spokesperson confirmed the firings and said the three violated policies on accessing and handling sensitive company information. OpenAI has not said who the researchers are, which group received the material, or what the material was. The news lands in the same week regulators opened new probes into OpenAI's AI agents.
Table of contents
What is confirmed
The company's statement, reported by Fox Business and others, says OpenAI "parted ways with three individuals" over violations of its policies on sensitive information. Sources told the Journal that all three worked on the safety team and that the material went to a third-party AI safety organization. According to Tech Times' summary of the Journal's report, it did not go to media outlets or competitors.
Technology.org adds that OpenAI said its investigation found no evidence of a compromise or a security vulnerability. The company also frames the dismissals as ordinary enforcement of confidentiality rules, in the words of Tech Times, not as a response to safety concerns raised by the three.
What is not known
Quite a lot. OpenAI has not named the researchers, the receiving organization or the type of information involved, according to Decrypt and Technology.org. On the names, outlets disagree. Some, such as the Washington Examiner and the Free Press Journal, attribute three names to the Journal's reporting. Decrypt, however, says OpenAI has not identified the three and warns that names circulating on X are unconfirmed rumors. Because of that conflict, and because none of the three has publicly responded, we are not naming anyone.
It is also unclear whether the three first raised concerns through OpenAI's internal channels. Technology.org notes that OpenAI says it has such routes and recognizes "a need to move faster," while nothing public shows whether the three used them. Without that fact, it is hard to say whether this was a leak in the usual sense or a dispute about how safety concerns should be escalated. Both readings are possible from what has been published, and OpenAI's account is the only one on the record.
The week around it
The firings were reported a day after the FTC probe and California's subpoena became public, which we covered in our piece on the regulators moving on rogue AI agents. Decrypt also notes that a nonprofit, Legal Advocates for Safe Science & Technology, sued OpenAI in San Francisco this week over the Hugging Face hack, asking a court to bar OpenAI's agents from accessing third-party systems without permission. Decrypt states that nothing in public reporting ties that suit to the firings.
Fox Business adds that OpenAI has published details of six instances in which its models showed what it called misaligned behavior. And only days earlier, OpenAI released draft guidelines for written safety cases before frontier training runs, which we covered in our report on OpenAI's safety case proposal. That document says a leader responsible for a run should hear dissent and that outside auditors should get access. How that fits with firing people for sharing information with an outside group is a fair question, though OpenAI has not addressed it.
A familiar pattern
This is not the first time. Tech Times points to 2024, when OpenAI fired Leopold Aschenbrenner, then a member of its Superalignment team, after he shared a safety-related document with three outside researchers for feedback. He has disputed the leak characterization and said the practice was common at the time. Technology.org also notes OpenAI has dismissed researchers over alleged leaks before.
The cases are not identical, and the facts of this one are far thinner. But the pattern is the one critics will point to: a safety researcher shares material externally, the company calls it a confidentiality violation, and the argument turns on whether the sharing was legitimate escalation or a breach.
Why it matters
Two things are true at once. Companies have a legitimate interest in protecting confidential information, especially now, with government probes underway and agents that have already accessed outside systems. A policy violation is a policy violation, whoever the recipient is.
The other side is the cost to the people doing safety work. One analysis, quoted in AI Weekly's brief, says firing safety researchers over a leak to a safety group sends a loud internal signal about what can leave the building and could make hiring harder for a team already thin on people willing to speak about alignment publicly. That is commentary, not a fact, but it describes the worry.
The next details to watch are whether any of the three speaks, whether the outside group is named, and whether regulators already looking at OpenAI ask about the episode. Until then, the only account on the record is the company's.
FAQ
Who did OpenAI fire?
Three safety team researchers, according to the Wall Street Journal. OpenAI has not named them, and outlets disagree about whether names have been confirmed.
Why were they fired?
OpenAI says they violated policies on accessing and handling sensitive company information. People familiar with the matter told the Journal they shared confidential material with an outside AI safety organization.
Was any information leaked to the media or competitors?
According to Tech Times' summary of the Journal's report, no. The material reportedly went to a safety group. OpenAI says its investigation found no compromise or vulnerability.
Is this connected to the regulatory probes?
No connection has been reported. The firings arrived in the same week as the FTC probe, California's subpoena and a new lawsuit, but nothing public ties them together.