Three researchers fired by OpenAI deny the alleged misconduct

Jasmine Wang, Tomek Korbak and Mikita Balesni, three safety researchers OpenAI fired last week, released an open letter on Oct. 8. In it they reject having handled sensitive data outside the company's procedures, according to TechCrunch. It is addressed to three internal OpenAI bodies: the Safety and Security Committee, the Safety Advisory Group and the Mission Advisory Council.

OpenAI initially said the firings were because they shared confidential information with an outside AI safety organization, following a Wall Street Journal report, according to Cybernews. This outlet covered that first report on Oct. 1, without names.

Now an OpenAI spokesperson told TechCrunch that the investigation found a "pattern of misconduct", and that it goes beyond contact with an outside evaluation group. The company did not say which policies were violated, and it has not formally responded to the letter.

The three give their own accounts:

• Wang: says she was delegated an executive's email for recruiting work, that IT did not remove it when she asked and that she reported within minutes a message she opened by mistake.
• Korbak: says he believed he was following the rules when he contacted outside evaluators about the Hugging Face incident, a case the letter calls "unprecedented".
• Balesni: says he kept his reporting line informed and removed sensitive details before sharing materials.

The three also deny leaking to The Information details about "less monitorable" architectures. In the letter's view, the firings are "chilling the open culture" OpenAI once valued. The signatories ask for outside safety auditors to be built in and for dialogue to continue with those who study the issue.

OpenAI shared with TechCrunch an internal memo from a research leader that acknowledges the researchers' contributions and says: "We do not terminate employees for raising concerns." According to the outlet, the memo says it agrees with the letter's recommendations.

The accusations from both sides remain unverified. This newsroom could not open the letter or the Journal report, so everything comes from TechCrunch and Cybernews.