Three safety researchers fired by OpenAI last week, Jasmine Wang, Tomek Korbak and Mikita Balesni, published a joint open letter on October 8. They deny the company's accusations and warn that the firings have left colleagues still at the company afraid to speak up. A day later, OpenAI responded in a statement on X, saying the decision to dismiss them stands and that its internal investigation found problems more serious than the letter suggests.
What each of the three was told
The letter is titled "OpenAI cannot make AI safe on its own" and is addressed to OpenAI's Safety and Security Committee, its safety advisory group and its mission advisory committee. The most concrete part is where each of the three describes the reason they were given for being dismissed.
Korbak says he was told only verbally that the reason was how he had communicated with the outside evaluation organization METR. He was given no details about what he said or did, or when, and nothing in writing. He argues that talking to METR was part of his job in the first place. METR previously worked with OpenAI on the investigation into the incident in which an agent broke into Hugging Face.
Balesni says that in his exit conversation, the company told him it no longer trusted him because he had been in too much contact with third-party safety organizations, implying that he had leaked intellectual property. He says he never shared the company's intellectual property, and that the work in question had been coordinated with his direct manager, the head of research and the board. He wrote on X that he believes the three were fired because they put safety ahead of the company's short-term interests.
Wang was told she was dismissed for accessing an executive's mailbox. Her account is that the access had originally been granted for recruiting work, that she asked IT to revoke it and the request was never carried out, and that after accidentally opening a sensitive email she told the executive within minutes. She has asked the company for a complete written list of the allegations.
The three also deny being the source of a leak. Media outlets previously reported on internal OpenAI discussions about "model architectures that are harder to monitor," and the letter states plainly: "We were not the source of the leak."
OpenAI's response
OpenAI's statement came from its head of research; most reports do not name the individual. It makes four main points: the investigation confirmed that the three violated clear policies on handling sensitive information; the breaches of trust it found go beyond what the open letter describes; personnel details are not disclosed, as is customary; and the dismissals had nothing to do with their raising safety concerns.
"We have not and do not terminate any of our employees for raising concerns."
The statement also contains points of agreement. OpenAI says it shares the letter's view that keeping frontier models monitorable requires investment from the whole industry, OpenAI included, and that it is finalizing contracts with third-party safety evaluators, with details to be announced in the coming weeks. Which organizations are involved, and whether the contracts will allow researchers to talk directly with outside evaluators, is not yet clear.
Connecting the dismissals of the past two years
This is not the first time OpenAI has fired researchers over alleged leaks. In April 2024, Leopold Aschenbrenner and Pavel Izmailov of the superalignment team were dismissed on suspicion of leaking information. Aschenbrenner publicly disputed it afterward, saying he had been targeted for raising safety concerns internally.
The timeline in this round is much tighter. On October 1, The Wall Street Journal reported that the three had been fired. On October 3, David Robinson, who is responsible for system cards, wrote in The Atlantic about why he left. The open letter followed on October 8, and OpenAI responded on October 9.
Compared with 2024, the focus of the dispute has shifted from whether anything was leaked to where the boundary lies between lab researchers and outside evaluators. The letter says internal policy was still being "worked out as we went," while OpenAI says the policy was clear. Both sides agree that independent evaluation matters; they disagree over who crossed the line.
Balesni worries that the company will use the dismissals to cut off its relationship with METR. No response from METR has appeared in public reporting so far, and OpenAI has not disclosed what the violations actually were. Until the third-party evaluation contracts are announced, all outsiders have to go on is each side's own account.
Sources: OpenAI official statement, TechCrunch, CocoLoop, Newsweek (full text of the open letter), Reuters, AFP; the three researchers' open letter and personal social media accounts were checked against their own statements, and CNBC was used to check the quotes in the statement.