OpenAI Asked 3 Researchers to Unlock Phones, Then Fired Them

OpenAI asked three of its safety researchers to unlock their personal phones during a leak investigation, then fired them a week later, the Wall Street Journal reported on 22 October 2026. “Last month, OpenAI called in the three outspoken safety researchers and asked them to unlock their personal phones,” Journal reporter Keach Hagey wrote on X. “A week later, at 2 p.m., they were all fired.”

Hagey wrote that the company was conducting a leak investigation into a story in The Information about Astra’s use of opaque architectures. We could not read the Journal’s article itself; this piece uses what the Journal and its reporters posted, the summary Techmeme published, and what the researchers and OpenAI have said on the record.

Business Pill · LEAST PRIVILEGE

A short Business Engineer explainer on least privilege: giving each person or system only the access it needs. It teaches the general idea only and says nothing about any company or person in this piece.

What the Journal Reported

• The phones. OpenAI called in the three researchers and asked them to unlock their personal phones, Hagey wrote; a week later, at 2 p.m., they were all fired.
• The chats. “Then, OpenAI asked to see their Signal chats. A week later, they were let go,” co-author Maxwell Zeff wrote on X, adding that the three “spent the summer fighting to keep AI’s chain-of-thought process visible.”
• The allegation. “OpenAI suspected the three fired researchers of leaking internal model designs to AI safety researcher Ryan Greenblatt, including via disappearing Signal chats,” according to Techmeme’s summary of the Journal’s article. That is the Journal’s account of what the company suspected; it is not a finding.
• The frame. The firings “came in the throes of a battle over how to keep AI systems under human control—and who gets to see the secret designs that could determine the winners of the biggest commercial race in history,” the Journal wrote in its post on X.

Greenblatt announced a move to METR, the outside evaluator, on X on 27 September: “I’m joining METR to work on more investigations like our Hugging Face report.” We did not find a public response from Greenblatt to the Journal’s report on Greenblatt’s X account or LessWrong profile as of 04:30 UTC on 23 October. A summary of the Journal’s article posted on X by AI commentator Rohan Paul at 06:04 UTC on 23 October says Greenblatt denies being the leaker; we have not read the Journal’s article to confirm the wording.

What Was Posted on X

Mikita Balesni, one of the three researchers, quoted the Journal’s post on 23 October. Balesni writes that OpenAI continues to not be able to say anything that makes sense to Balesni about the firings, and cites what Balesni describes as OpenAI claiming to journalists that Balesni was not on the Hugging Face investigation. It matters because it is one of the three researchers responding directly to the Journal’s new reporting.

Jasmine Wang, quoting the Journal’s Maxwell Zeff the same day, compares AI with nuclear and finance, where Wang writes that employees have defined channels for raising safety concerns with auditors and regulators. Wang argues AI needs similarly clear rules governing relationships between labs, outside evaluators and employees, which is the policy question this case raises.

Tomek Korbak’s post of 8 October gives Korbak’s account: Korbak writes of being OpenAI’s main technical point of contact with METR, of being told verbally that the firing was because of the way Korbak communicated with METR, and that nothing was put in writing. It is the source of the quote card in this piece.

What OpenAI Has Said

“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 to the Journal on 1 October, TechCrunch reported. The report at the time did not name the researchers, the organization or the information involved, TechCrunch said.

Our 8 October piece reported a memo OpenAI shared with TechCrunch, attributed to a research leader: “I want to be very clear that these decisions were not about raising safety concerns or speaking out.”

The Data Point: Four Accounts of One Decision

Set side by side, the on-record explanations do not describe the same act. OpenAI cites the handling of sensitive information; two of the researchers say they were each given one, different reason; the third says what Balesni believes.

Each account is as its author gave it; none has been independently verified. Korbak and Wang describe the reason they say they were given; Balesni describes a belief.

What the Researchers Say Now

Balesni quoted the Journal’s post on 23 October, writing that OpenAI “continues to not be able to say anything that makes sense to me about our firings” and keeps acting adversarially, citing what Balesni described as OpenAI “claiming to journalists I wasn’t on the Hugging Face investigation, while there’re a number of people who could confirm otherwise.”

Wang, quoting Zeff’s post the same day, wrote: “OpenAI says they support embedded auditing but then fires employees for talking to those same auditors.” Wang argued that “AI needs similarly clear rules governing relationships between labs, outside evaluators, and employees,” comparing AI with nuclear and finance.

Korbak, who wrote of being OpenAI’s main technical point of contact with METR, wrote on 8 October: “To be clear, talking to METR was my job.”

Connecting the Coverage

Our open letter piece reported on 8 October that the three researchers wrote that they were not the source of a leak to The Information about less monitorable architectures. What’s new: the Journal now reports that the leak investigation reached their personal phones and Signal chats, and names the outside researcher OpenAI suspected.

Our METR piece reported that METR and Redwood Research published a write-up on 22 October 2026 of a three-week exercise inside OpenAI, estimating that on the order of 10% of eligible internal Codex traffic was not monitored between May and July 2026. That write-up came out the same day as the Journal’s report.

Our recorder piece reported that the US Justice Department said on 13 October 2026 that an OpenAI engineer had been charged with using or endeavoring to use a device to intercept oral communications; that is a separate case. Our Madry piece reported that Aleksander Madry, who once led OpenAI’s Preparedness team, is returning to OpenAI, as both Madry and chief scientist Jakub Pachocki wrote on X on 19 October 2026.

The Structural Read

As we read it, the dispute is about where an employee’s job ends: Korbak wrote that talking to METR was Korbak’s job, while OpenAI’s statement says the three mishandled sensitive information outside established company procedures.

As we read it, the Journal’s reporting adds a named outside researcher to the company’s suspicion, according to Techmeme’s summary; neither the Journal’s posts nor OpenAI’s statement, as we read them, says what was shared.

As we read it, Wang’s post of 23 October turns the case into a question about rules: Wang wrote that AI needs clear rules governing relationships between labs, outside evaluators and employees.

Three Implications

FOR OPENAI Its on-record explanation remains the 1 October statement about mishandled sensitive information; Balesni wrote on 23 October that the company continues not to say anything that makes sense to Balesni about the firings.

FOR OUTSIDE EVALUATORS Greenblatt announced a move to METR on 27 September; the Journal reports OpenAI suspected leaks to Greenblatt, which this piece reports as a suspicion, not a finding.

FOR AI SAFETY STAFF Balesni wrote on 8 October that former colleagues worry their personal phones will be searched for messages to the three and to third parties.

The Business Engineer Lens

This story maps onto the Business Engineer framework Inside Anthropic’s Permission Layer.

The framework’s starting point: “The governance layer sits at the intersection of capability, regulation, national security, access control, and geopolitical alignment.”

As we read it, this case sits in the access-control part of that layer inside one lab: who may see a model’s design, and which outside evaluators staff may talk to. The framework was published in June 2026 and does not discuss OpenAI’s firings.

What Is Not Established

We did not read the Journal’s article, which is behind a captcha here; its findings appear in this piece only as its reporters and Techmeme summarised them. We have not seen the Signal chats, the internal investigation or any document showing what, if anything, was shared, and with whom. The suspicion that the researchers leaked designs to Greenblatt is OpenAI’s, as the Journal reported it; the researchers have said they were not the source of the leak to The Information. We did not contact OpenAI, the researchers, Greenblatt or METR. Nothing here is investment advice.

The Bottom Line

The Journal reported that OpenAI asked three safety researchers to unlock their personal phones during a leak investigation and fired them a week later, and that the company suspected them of leaking internal model designs to Ryan Greenblatt. OpenAI says they mishandled sensitive information; each researcher gives a different account of why they were let go.