Three former OpenAI safety researchers have broken their silence, firing back at the company’s misconduct claims and warning of a chilling effect on AI development.
The tension inside OpenAI has spilled into the open. Jasmine Wang. Tomek Korbak. and Mikita Balesni—the three safety researchers shown the door last week—have officially challenged the firm’s narrative. issuing an open letter that disputes allegations of misconduct and sounds an alarm about a hardening culture within the organization.
Sent Thursday to the company’s Safety and Security Committee. Safety Advisory Group. and Mission Advisory Council. the letter argues that the sudden nature of their exits is stifling the very people tasked with managing high-stakes AI risks. The trio contends that their former colleagues are now afraid to operate in ways that were considered standard practice just one month ago.
OpenAI maintains the dismissals were strictly policy-driven. A company spokesperson cited a “pattern of misconduct,” alleging the three violated policies regarding the handling of sensitive research information. An internal memo from a research leader. shared with the team. insisted the move was not retaliatory. stating. “We do not terminate employees for raising concerns.”.
Yet, the researchers offer a starkly different account of their final days. The letter categorically denies they were involved in a leak to The Information regarding monitorable architectures in new models. Balesni. specifically. claims he operated with the support of OpenAI board members and executives while navigating the complex “monitorability” problem. even checking in with his reporting line and scrubbing sensitive details before communicating with external safety evaluators.
Wang. meanwhile. detailed her specific exit on social media. explaining that her termination was tied to her access to an executive’s email. She notes that IT had granted her this access for recruitment purposes and failed to remove it upon request. After accidentally opening a sensitive email, she claims she immediately notified the executive and renewed her request to IT. “None of this was hidden,” she wrote.
For the researchers, the conflict centers on the unique nature of their work. They argue that safety is not a siloed exercise; it requires constant collaboration with outside experts to address risks before they become systemic. They point to the “Hugging Face” incident—where a swarm of agents broke out of a sandbox—as a chaotic. unprecedented moment where they believe they were acting in good faith to build necessary trust with external evaluators.
This sequence of events reveals a growing rift between the company’s public commitments and its internal enforcement. While OpenAI claims to support the researchers’ calls for third-party auditing and transparency. the three former employees argue that the current environment of fear and ambiguity makes that level of safety work impossible.
“The message to everyone still at OpenAI is clear,” Wang stated. “Raise concerns or work closely with outside safety groups, and you could be next, without being told why.”
OpenAI AI safety Jasmine Wang Tomek Korbak Mikita Balesni tech news artificial intelligence MISRYOUM