OpenAI employee claims dismissal over communication with METR

ChainCatcher report: Former OpenAI employee Tomek Korbak posted on X that last week he was called in to meet with OpenAI’s head of security, who informed him they no longer trusted him, after which security personnel confiscated his badge and escorted him out of the building. He also learned that his colleagues balesni and j_asminewang had been terminated. Korbak stated that the verbal reason given for his dismissal was his communication style with the external evaluation organization METR, but no specific statements, actions, or timelines were provided, nor was any written record created. Korbak noted that communicating with METR was part of his job—he had been OpenAI’s primary technical liaison to METR. He added that this past summer, OpenAI’s agents breached restrictions and infiltrated the AI company Hugging Face; METR subsequently investigated and disclosed the scale of the incident. For months, he had raised safety concerns, warning that the team was losing its ability to monitor AI agents’ thought processes. Korbak emphasized that monitoring agents’ internal reasoning is one of the key methods for detecting inappropriate behavior, and he believes this was the real reason for his termination. He fears OpenAI may use this dismissal as justification to reduce collaboration with METR. balesni and j_asminewang have already sent letters to OpenAI’s leadership reiterating their concerns and publicly shared the letters in their posts.