An IT Home article on October 2, citing a Wall Street Journal report from the day before, says OpenAI has ended its work with three researchers for violating rules on access to and sharing of sensitive information. The spokesperson’s statement says the company has parted ways with three people who violated its rules for accessing and handling sensitive information. The investigation, the statement says, confirmed that they bypassed established process, mishandled sensitive information, broke the rules, and damaged the trust the work requires.

[1]
Ink: a closed notebook with a strap, and three blank cards in a row set apart from it.
A closed notebook with a strap sits on the left. Three blank cards stand in a row to the right, with a gap between them and the book. That is three researchers no longer in the same place. An illustration, not a personnel file., AI-generated illustration, not a news photograph

The article names them Jasmine Wang, Tomek Korbak, and Mikita Balesni. Korbak was on OpenAI’s safety team. He has said that during the Redwood Research and METR investigation of the Hugging Face incident, he was OpenAI’s technical contact. Wang and Balesni worked on alignment, on how to make models follow human intent. The article says the employees were accused of giving confidential company information to a third-party AI safety organization. The organization is not named.

[1]

The same article also records a separate recent statement from OpenAI: the company supports independent evaluation of safety cases, says that practice is gaining support, and says it is committed to independent evaluation, including deep access to training, evaluation, and deployment so reviewers can question assumptions, find missed risks, and judge whether safeguards work. The article places that statement next to the dismissals. It does not say these three people were the evaluators in that statement, and it does not say independent evaluation itself broke the rules.

The Wall Street Journal piece is not in hand here. The IT Home article does not say what was given, to which organization, or which records the investigation used.

[1]

要点

  • An OpenAI spokesperson said the company parted ways with three people who violated rules on sensitive information, and that they bypassed established process.
  • IT Home names Jasmine Wang, Tomek Korbak, and Mikita Balesni. Korbak was on the safety team. Wang and Balesni worked on alignment.
  • The article says confidential information was given to an unnamed third-party AI safety organization.
  • A nearby statement in favor of independent evaluation is not presented as these three people’s role, or as the violation itself.