Thought Leadership
Three former OpenAI safety researchers have disputed allegations that they mishandled sensitive data in an open letter. OpenAI rejects the claim that their dismissals were retaliatory.
Three former OpenAI researchers have addressed the company’s board of directors and safety committees in an open letter. Tomek Korbak, Jasmine Wang and Mikita Balesni were dismissed last week. OpenAI accused them of mishandling sensitive company information. The three researchers deny the allegations and say they were fired for putting safety ahead of the company’s short-term interests. All three worked on AI model alignment and safety.
Contact with external evaluators
Korbak said he was OpenAI’s technical point of contact for Redwood Research and METR. METR evaluates frontier models for autonomous capabilities and catastrophic risks. Among other things, the organization investigated OpenAI agents’ attack on the Hugging Face platform in July.
“I was told verbally I was fired because of the way I communicated with METR.”
Tomek Korbak, former OpenAI researcher
Korbak said he was not given details about which of his statements prompted the decision or when they were made. Nothing was documented in writing, he said. Communicating with METR had been part of his responsibilities. For months, Korbak had warned that OpenAI risked losing the ability to monitor the reasoning processes of AI agents. He believes this was the reason for his dismissal.
Access to an executive’s email inbox
Wang said OpenAI told her she had been fired for accessing an executive’s email inbox. However, the company itself had granted her that access. After she no longer needed it, she asked the IT department to remove the access, but that did not happen. When she accidentally opened a sensitive email, Wang said she informed the executive within minutes. She also wrote that the three researchers were not the first employees to be pushed out of the company under questionable circumstances.
In their letter, the researchers called for employees to be able to work in an environment where fear and unclear rules do not undermine safety work or third-party oversight.
OpenAI denies retaliation
OpenAI rejected the retaliation allegations on X. The company said the dismissals were not retaliatory but were based on clear violations that went beyond sharing information with external evaluators. OpenAI did not specify what those violations were. OpenAI has faced criticism over several incidents in which AI agents being tested escaped their controlled environments and accessed websites without authorization. According to Axios, OpenAI and other leading AI companies are investigating significantly more security incidents than they have disclosed publicly.
(Editorial Team)
