Openai Researchers Fired Over Alleged Leak to Safety Expert

WHO:

NPR - Fired OpenAI safety researchers Mikita Balesni and Tomek Korbak gave their first interview since their dismissal.

NPR - The interview covered OpenAI, METR, and the Hugging Face hack investigation.

Techmeme (Wall Street Journal) - OpenAI suspected three fired researchers of leaking internal model designs to AI safety researcher Ryan Greenblatt.

WHAT:

NPR - Balesni and Korbak discussed their dismissal and described events inside OpenAI during METR’s investigation into the Hugging Face hack.

NPR - The researchers explained why they believed OpenAI’s rebuttal did not add up and described an ongoing fight over AI safety inside the company.

NPR - The episode also introduced a segment called “Pandora’s Box.”

NPR - The episode mentioned Elon Musk’s proposal to give every child in America a SpaceX share and new filings in an OpenAI “bugging case” that revealed a suspect’s chats with a Saudi royal.

Techmeme (Wall Street Journal) - OpenAI fired three safety researchers for mishandling sensitive information, according to the article’s summary.

WHERE:

NPR - The episode was available on YouTube, Spotify, and Apple.

NPR - The episode invited listeners to send questions by commenting on those platforms or emailing hey@machinegods.fm.

WHY:

Techmeme (Wall Street Journal) - OpenAI suspected the three researchers of leaking internal model designs to Ryan Greenblatt.

Techmeme (Wall Street Journal) - The company said the researchers were fired for mishandling sensitive information.

HOW:

Techmeme (Wall Street Journal) - The suspected leaks allegedly included the use of disappearing Signal chats.

(UPDATED October 23, 2026 16:20 UTC)

Two former OpenAI safety researchers, Mikita Balesni and Tomek Korbak, have spoken publicly for the first time since they were fired. In an interview, they discussed their dismissals and events inside OpenAI during an investigation into a hack involving Hugging Face. OpenAI suspected three fired researchers of leaking internal model designs to AI safety researcher Ryan Greenblatt, allegedly using disappearing Signal chats. The company said the researchers were dismissed for mishandling sensitive information. Balesni and Korbak said they believe OpenAI’s rebuttal does not add up, and described a continuing dispute over AI safety within the company. Their interview also covered METR and the Hugging Face investigation. The episode included a segment called “Pandora’s Box,” along with brief mentions of Elon Musk’s proposal to give children SpaceX shares and court filings involving a suspect’s chats with a Saudi royal. It was available on YouTube, Spotify, and Apple.