Artificial Intelligence
OpenAI fires three safety researchers after internal review
OpenAI dismissed three safety researchers linked to a Hugging Face investigation, and the researchers later warned in an open letter that the terminations had frightened staff and left rules unclear.
- OpenAI fired Tomek Korbak, Jasmine Wang and Mikita Balesni, and the three later sent an open letter to OpenAI's Safety and Security Committee, Safety Advisory Group and Mission Advisory Council.
- Korbak said he was told verbally that his communication with METR was the reason, and he said no written explanation was given.
- The three said they were not the source of a leak to The Information and said Korbak and Balesni had worked on the Hugging Face investigation.
- OpenAI said an internal investigation found that the three had breached policies on sensitive information and said the firings were not for raising safety concerns.
OpenAI fired Tomek Korbak, Jasmine Wang and Mikita Balesni after an internal review, according to the company and the researchers. The three then wrote to OpenAI's Safety and Security Committee, Safety Advisory Group and Mission Advisory Council.
Korbak said he was told in person that the issue was his work with METR, the external safety lab that reviewed the Hugging Face incident. He said he had been OpenAI's main technical contact for that work.
The three said they were not the source of a leak to The Information about less monitorable systems. They also said Wang had opened a sensitive email after access to an executive inbox was not removed.
OpenAI said its investigation found policy breaches over sensitive information and denied that the dismissals were retaliation for safety concerns. It also said it was working on contracts with external safety auditors and on industry-wide commitments to model monitorability.
Named in this story
People
- Tomek Korbak
- one of the fired safety researchers and a signatory of the letter
- Jasmine Wang
- one of the fired safety researchers
- Mikita Balesni
- one of the fired safety researchers
- Jan Leike
- OpenAI's former head of super AI safety who left for Anthropic after criticising safety culture
Companies
- OpenAI
- fired the three researchers and issued the response
- Hugging Face
- its platform was involved in the incident under discussion
- Anthropic
- hired Jan Leike after he left OpenAI
Organisations
- METR
- the external safety lab that examined the incident
- OpenAI's Safety and Security Committee
- received the open letter from the researchers
- OpenAI's Safety Advisory Group
- received the open letter from the researchers
- OpenAI's Mission Advisory Council
- received the open letter from the researchers
How the source tells it
The register was adversarial and crisis-led, with blame-heavy language and repeated high-stakes claims about safety, fear and company conduct.
- alarm crisis framing and worst-case language around unsafe AI systems
- outrage pejorative wording and repeated fault-finding aimed at OpenAI
- speculation motives and consequences were inferred with more certainty than the evidence shown