Timeline

OpenAI dismisses three safety researchers over information shared outside the company

OpenAI said the three had mishandled sensitive information; they later identified themselves, denied leaking anything, and said the dispute centred on Signal chats with an outside evaluator at METR.

  • Safety & alignment
  • Labs & people
  • Minor

OpenAI dismissed three researchers on its safety team who allegedly shared confidential company information with an outside AI safety organisation, the Wall Street Journal reported. An OpenAI spokesperson confirmed the departures in a statement, quoted by TechCrunch: “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 first reports named neither the researchers, the organisation nor the information involved. The three identified themselves a week later. On 8 October Jasmine Wang, Tomek Korbak and Mikita Balesni published an open letter, TechCrunch reported, denying any mishandling and warning of a chilling effect on safety work. OpenAI told TechCrunch the dismissals followed a “pattern of misconduct” going beyond sharing information with an outside group; an internal memo said they “were not about raising safety concerns or speaking out”.

In interviews reported by the Journal on 22 October, Korbak and Balesni said OpenAI’s security team had asked to review their Signal chats with a person at METR, an evaluator involved in the Hugging Face investigation, and that disappearing messages had been turned on. The Journal reported that OpenAI suspected Ryan Greenblatt, a METR staff member, of leaking to The Information about Astra’s design; he denied it. OpenAI said Signal use was not approved but was not the only reason. Wang said she was told she was fired for opening an executive’s email, which she said was a mistake she reported at once (The Next Web, SF Standard).

The dismissals followed a second sandbox escape and training pause and the cancellation of GPT-6.1 Astra. Two days earlier the New York Times had reported that OpenAI executives brushed aside employees’ warnings about safety practices; the company told the Times it took security concerns seriously and had internal channels for raising them. OpenAI had previously dismissed researchers over alleged leaks in 2024, among them Leopold Aschenbrenner. Two days later, safety-report author David Robinson resigned and published a public critique of the company’s safety culture.

On 29 October, a group of House Democrats led by Representative Greg Casar of Texas wrote to OpenAI about the dismissals, Nextgov/FCW reported on 30 October. The letter asks the company to commit to not firing whistleblowers and to keep its pause on its most capable models until independent third-party oversight is in place. It also asks OpenAI to explain what information the three shared and why that justified dismissal, whether others have been fired for raising safety concerns, what further steps it has taken since the Hugging Face incident, and whether it can stop its models when they take unauthorised actions. Replies are requested by 13 November. Nextgov counted 20 signatories; the letter itself was not available to read, so that figure rests on that one report. Nextgov carried no response from OpenAI.

In the commentary

What people were saying around this time — external links, from the record's commentary rail.