OpenAI fired three safety researchers a week after a review of their Signal chats: what the WSJ reported, and what each side says

The three OpenAI safety researchers whose firing became public on October 1, 2026, were dismissed after the company's security staff found disappearing messages in their Signal chats with an outside AI safety researcher, The Wall Street Journal reported on October 22–23, 2026. OpenAI had already confirmed on October 1 that it had parted ways with three people for "violating our policies on accessing and handling sensitive company information." The researchers, Tomek Korbak, Mikita Balesni and Jasmine Wang, say they weren't the source of the leak OpenAI was hunting. The leak in question was a report about how monitorable OpenAI's newest model is, covered in our loop transformers explainer.

This story is based on OpenAI's statements as quoted by TechCrunch, Fortune and the WSJ, the researchers' open letter, Balesni's and Korbak's on-the-record interview on NPR's Machine Gods podcast (October 23, 2026), the WSJ's October report as republished in summary by the Benton Institute and described by The San Francisco Standard, and reporting by TechCrunch and Fortune, checked October 25, 2026. 9to5AI couldn't read the WSJ article itself.

The short version

• Who: Tomek Korbak, Mikita Balesni and Jasmine Wang, safety and alignment researchers at OpenAI.
• OpenAI's position: the three violated policies on handling sensitive information, and "these decisions were not about raising safety concerns or speaking out."
• What the WSJ added: security staff asked to see the researchers' phones in September and found disappearing Signal messages with an outside safety researcher; a week later the three were dismissed. Balesni and Korbak confirmed the Signal check and the disappearing messages on NPR's Machine Gods.
• The researchers' position: they weren't the source of the leak, and they acted within norms that were still being written.
• Not established: that any of the three leaked anything. No outlet has reported evidence that they did.

What did the Wall Street Journal report?

In an article by Maxwell Zeff, Keach Hagey and Berber Jin, the Journal reported that in September, OpenAI's security staff met with the three researchers and asked them to unlock their personal phones. The company suspected they had shared details of its internal model designs with an outside AI safety researcher. Security officers found that some of them had turned on disappearing messages in their Signal chats with that person. A week later, the three were called into simultaneous 2 p.m. meetings and dismissed. That account comes from the Benton Institute's republished summary of the article.

Korbak told the Journal: "I was told that there was a breach of trust because I interacted with a third party through unsanctioned channels."

Balesni and Korbak gave the same account on the record in NPR's Machine Gods podcast, released October 23, 2026. Balesni said that a week before they were terminated, they were pulled into a security investigation into a leak of OpenAI IP to journalists, tied to The Information's article on less monitorable architectures. The security team asked to see their Signal messages with an outside safety researcher on their personal phones: "We all did that," Balesni said. Korbak said they both had disappearing messages turned on, which he called a security best practice. NPR said Machine Gods contacted OpenAI and didn't hear back.

The San Francisco Standard, describing the Journal's reporting, said the review was part of OpenAI's search for the source of an article in The Information, which reported that OpenAI's newest model, Astra, uses a technique that reveals less of its reasoning and could be harder to monitor. According to the Standard, Balesni told the Journal that OpenAI suspected Ryan Greenblatt, formerly chief scientist at Redwood Research and now at METR, the Berkeley evaluation nonprofit. Greenblatt denied being the source.

Per the Journal, as described by the Standard, OpenAI said it hadn't approved employees' use of Signal, but that this wasn't the only reason for the firings.

What has OpenAI said?

OpenAI's statements, as quoted by others, are:

• To the WSJ, around October 1: "We have parted ways with three individuals for violating our policies on accessing and handling sensitive company information." And: "Our investigation confirmed that these individuals mishandled sensitive information outside established company procedures." (TechCrunch, October 1)
• In an internal memo shared with TechCrunch: "I want to be very clear that these decisions were not about raising safety concerns or speaking out." A spokesperson said an investigation found a "pattern of misconduct." (TechCrunch, October 8)
• On X, after the researchers' letter: "Our internal investigation uncovered a significant breach of trust beyond what's outlined in the letter they published." (Fortune, October 9)

OpenAI hasn't said which policies were broken or what information was involved. Bloomberg reported that at least some of it concerned the architecture of OpenAI's infrastructure, according to Fortune. OpenAI didn't respond to Fortune's or The San Francisco Standard's requests for comment.

What do the fired researchers say?

The three signed an open letter to OpenAI's Safety and Security Committee, Safety Advisory Group and Mission Advisory Council. It says:

• On the leak: "We were not the source of the leak for The Information article about supposed new, less monitorable architectures." They say they don't know who was, and that the article undercut their own work.
• Korbak was OpenAI's technical contact for METR's investigation into OpenAI agents that hacked Hugging Face, and the letter says he worked closely with outside counterparts within the policies then in place. Our rogue agents explainer covers that incident.
• Balesni was coordinating cross-company monitorability work and checked in with his reporting line, the letter says.
• Wang, according to the letter and TechCrunch's account of her post on X, had been given delegated access to an executive's email for recruiting, asked for it to be removed, accidentally opened a sensitive email and reported it to the executive within minutes. "None of this was hidden," she wrote, per TechCrunch.

They asked OpenAI to keep its commitment to embed third-party safety auditors: "Our firing should not be used as a pretext for stepping away from those partnerships."

On Machine Gods, Korbak said: "But somehow at OpenAI, when you are not able to prove your innocence, you are presumed guilty."

Is this whistleblowing?

That isn't settled, and the facts that would decide it haven't been made public. Rep. Greg Casar wrote that the firings "look like" the firing of whistleblowers, Fortune reported. Charlie Bullock, a senior policy director at LawAI, told Fortune that California's whistleblower protections generally cover disclosures to government, law enforcement or internal channels, not to private third parties or the press, so on what is known, the researchers likely wouldn't qualify. He said he was withholding judgment on whether the firings were right.

Have OpenAI safety staff left before?

Yes. TechCrunch, citing The Information, reported that OpenAI fired researchers Leopold Aschenbrenner and Pavel Izmailov in 2024 over alleged leaks. Fortune reported that David Robinson, who led the safety reports published with OpenAI's major launches, resigned in late September 2026 and wrote in The Atlantic that the company's culture is broken.

The firings also landed during a dispute over outside evaluators' access to OpenAI. On October 22, 2026, a METR employee and a Redwood Research employee published a red-team review of OpenAI's internal agent monitoring, covered in our red team story. Earlier, Reuters reported on labs seeking control over what evaluators publish; see our evaluator independence story.

What's confirmed, reported and uncertain

• Confirmed: OpenAI dismissed three people over its handling-of-information policies (OpenAI statements quoted by TechCrunch and Fortune); their names and their denials (their own letter).
• On the record from the researchers (NPR's Machine Gods): the security investigation a week before the firings, the request to see Signal messages on their personal phones, and that disappearing messages were on. OpenAI hasn't commented on these details publicly.
• Reported (WSJ, via the Benton Institute and The San Francisco Standard): the simultaneous 2 p.m. meetings, OpenAI's suspicion of Greenblatt, and OpenAI's statement that Signal wasn't the only reason.
• Uncertain: what information OpenAI believes was mishandled, whether anything was leaked by anyone among the three, and who the source of The Information's article was.