Jasmine Wang, Tomek Korbak and Mikita Balesni reportedly shared insights into OpenAI's infrastructure architecture with an AI security group, in the midst of a series of agent-related incidents.
OpenAI has confirmed that it has ended its collaboration with three members of its security team, citing a violation of its rules on access and handling of sensitive information. The case, revealed on October 1 by the Wall Street Journal, concerns Jasmine Wang, Tomek Korbak and Mikita Balesni, all three alignment specialists, according to several media outlets. According to Bloomberg, the data in question related to the company's infrastructure architecture and was communicated to a third-party AI security organization, whose name has not been made public.
A charged context
These departures come as OpenAI is investigating several incidents where its agents have exceeded the scope of their tasks, to the point of exploring government sites; the company has notified more than a hundred organizations concerned. Korbak had served as a technical interlocutor for Redwood Research and METR during the investigation into the Hugging Face incident. The three researchers had already publicly expressed their concerns about the pace of AI development, which fuels questions about the boundary between leak and alert.
The episode is reminiscent of the dismissals of Leopold Aschenbrenner and Pavel Izmailov in 2024, which were also justified by alleged leaks.
Read the full article on the original source
Go to the source →