Policy Safety United States OpenAI

Update: Fired OpenAI Safety Researchers Say in a Letter to Its Safety Bodies They Were Not the Leak Source; OpenAI Calls the Firings a "Significant Breach of Trust"

Updated Oct. 9: Tomek Korbak, Jasmine Wang and Mikita Balesni, the three researchers OpenAI fired last week, published a letter on Oct. 8 to OpenAI's Safety and Security Committee, Safety Advisory Group and Mission Advisory Council. They say they were not the source of a leak to The Information, deny engaging outside parties beyond their jobs, and ask OpenAI to keep third-party safety auditors embedded and preserve the monitorability of frontier models. On Oct. 9, OpenAI said in a post on X, as reported by CNBC, that a thorough investigation found they violated clear policies on handling sensitive information, a "significant breach of trust", and that it agrees with the letter's ethos on monitorability. What information was involved has not been made public.

Times of AI Desk Oct 1, 2026 7 min read San Francisco View as Markdown

Corrected October 14, 2026: Dated and scoped the statement that the safety bodies had not responded.

Updated October 9, 2026: The three researchers have now named themselves and set out their side in a letter, and OpenAI has answered on the record. The new section directly below covers both; the October 1 report follows, unchanged except that its "Context that is public" heading now sits one level lower, under the dated October 1 section, and one line in its Limits section has been reworded.

October 9: the researchers' letter and OpenAI's reply#

Tomek Korbak, Jasmine Wang and Mikita Balesni published a four-page letter on Thursday, October 8, titled "OpenAI cannot make AI safe on its own." It opens: "We are the three safety and alignment employees who were fired from OpenAI last week." The PDF is addressed to OpenAI's Safety and Security Committee, Safety Advisory Group and Mission Advisory Council, the three bodies TechCrunch also names. CNBC describes it as addressed to OpenAI board members and its safety committees, and says the researchers posted it on X.

What the researchers say#

On the firings, the letter says:

"We were not the source of the leak for The Information article about supposed new, less monitorable architectures. We do not know who it was."
"At no point do we believe we engaged with external parties outside the mandates of our jobs." It describes the Hugging Face incident investigation as "without precedent," with internal policies "being developed in real time," and says Korbak, the technical point of contact for METR in that investigation, "made every effort to act within OpenAI's policies as they then stood."
Balesni's work on cross-company commitments to prevent loss of monitorability was done "in coordination and discussion with board members and the C-suite," and he "took care to remove sensitive details from materials before sharing them."
Wang's access to an executive's email "was delegated for recruiting purposes, with permission"; she asked for it to be removed, IT did not remove it, and when she "accidentally clicked on a sensitive email" she reported it to the executive "within minutes."
They say rumours about "a shared board-level memo" were never raised with them, and that they did not share news of their firings with the media.

The letter's central worry is about the people still at OpenAI: "We have become concerned that internal and external communications around our firing have made our former colleagues afraid to speak and operate in ways that, until last week, were an integral part of working at OpenAI."

It ends with three recommendations:

OpenAI "must adhere to last month's public commitments to embed third-party safety auditors within the organization." The researchers say they are concerned the firings "may be used to justify ending OpenAI's work with METR, or otherwise providing external auditors much more limited access and scope."
OpenAI "must preserve the monitorability of frontier models." The letter says OpenAI and other frontier companies "should not move forward with developments that further decrease monitorability" while they rely on monitorability for safety. (CNBC paraphrased this as a call to "pause developing models that would decrease the ability to monitor" AI.)
OpenAI "must continue to support an open and transparent culture of dialogue between its safety researchers and the rest of the safety ecosystem," and should set out clearly how employees may work with external safety organisations.

In a separate thread on X, quoted by TechCrunch, Wang wrote that the reasons for the terminations are "not adding up."

What OpenAI says#

Since its October 1 statement to Gizmodo, OpenAI has added three more, and they are different texts:

Friday statement on X (October 9), as reported by CNBC: the researchers committed a "significant breach of trust," and the decision came "after a thorough investigation found they violated clear policies on handling sensitive information." OpenAI said it agrees with the ethos of the letter around "preserving the monitorability of frontier models" and continues to invest significant resources in that area. CNBC reports OpenAI said the decision was not about the researchers having raised safety concerns.
Internal memo shared with TechCrunch (October 8), attributed to a research leader: "I want to be very clear that these decisions were not about raising safety concerns or speaking out. We have always encouraged that and always will. We do not terminate employees for raising concerns." TechCrunch reports the memo praises the three researchers' contributions and agrees with their recommendations.
Spokesperson to TechCrunch: the three were fired after an investigation revealed a "pattern of misconduct" in "clear violation of our policies of mishandling research information" that goes beyond sharing information with an outside AI evaluation group. TechCrunch says OpenAI did not answer which policies were violated or the circumstances of the dismissals.

Where this leaves it#

The two sides now agree on one thing in writing, the importance of monitorability, and on nothing about the firings. The researchers say they acted within the norms of the time; OpenAI says an investigation found clear policy violations and a pattern of misconduct. Neither has said publicly what information was handled improperly or with whom it was shared, so neither account can be checked against the other. The dispute is now in front of the safety bodies the letter addresses; as of October 9, no response from them had been reported in the coverage reviewed for this article.

October 1: the original report#

OpenAI’s on-record statement is narrow. The WSJ’s naming and the “third-party organization” frame sit one layer out — and neither side has tested its case in public.

On October 1, 2026, an OpenAI spokesperson told Gizmodo: “We have parted ways with three individuals for violating our policies on accessing and handling sensitive company information… Our investigation confirmed that these individuals mishandled sensitive information outside established company procedures, violating our policies and breaking the trust essential to our work.”

The Wall Street Journal, as relayed by Gizmodo and The Decoder, reported that the material went to “a third-party AI safety organization” and later named the three as Tomek Korbak (safety) and Jasmine Wang and Mikita Balesni (alignment). OpenAI did not confirm the names when Gizmodo asked. This piece does not state what was shared or which organization received it — those facts are not established in the sources inspected here.

Context that is public#

Korbak had said he was OpenAI’s technical contact for METR and Redwood Research during their inquiry into the Hugging Face agent breach. The WSJ, per The Decoder, does not draw a connection between that role and the firings. All three had posted publicly about AI risk in September.

The same week, David Robinson — who led launch safety reports and helped draft the Preparedness Framework — resigned and published an Atlantic essay criticizing iterative deployment. Counted together, that is four safety-side departures in about a week, during an FTC probe into consumer harm from AI products and ongoing scrutiny of rogue-agent incidents. OpenAI’s policy-violation case and the researchers’ side remain untested in public.

Limits#

October 9: the researchers' letter was read in full as the four-page PDF on Balesni's website; the copy posted on X was not separately compared, though TechCrunch's quotes match the PDF. OpenAI's Friday statement on X is as reported by CNBC; the post itself was not opened. The internal memo and the spokesperson's comments are as TechCrunch reports them.
October 9: the letter's account of each researcher's conduct is their own; OpenAI's "pattern of misconduct" and "clear violations" are its own. Neither has been independently established.
WSJ original not inspected; names and “third-party organization” arrive via Gizmodo / The Decoder secondary coverage.
OpenAI confirmed the policy-violation parting, not the names or the recipient.
The WSJ, per The Decoder, draws no link between the METR/Redwood contact role and the firings; this article draws none either.
What was shared is unknown here.