The three safety researchers fired by OpenAI last week, Jasmine Wang, Tomek Korbach, and Mikita Baresny, have released an open letter denying the company’s claims that they deviated from established internal procedures and mishandled sensitive information, warning that their firings signaled a chilling effect that would have ripple effects throughout the company’s culture.
“We are concerned that internal and external communications regarding our termination have caused former colleagues to fear the words and actions that were an integral part of working at OpenAI until last week,” the researchers said in an open letter Thursday to OpenAI’s Safety and Security Committee, Safety Advisory Group, and Mission Advisory Council.
The researchers were fired last week for allegedly sharing confidential company information with a third-party AI safety organization. OpenAI said they violated its policies by “accessing and handling confidential company information.”
“AI is no ordinary technology, and OpenAI is no ordinary company,” Wang, Korbak, and Balesni wrote. “Those of us who work in safety rely on close collaboration with outside experts to recognize risks before anyone else and think about how to address them. Having the freedom to take risks without fear, and having clearly defined internal procedures that enable this work, is itself an essential safety mechanism.”
They said their firings represented a broader change in OpenAI’s culture, which had encouraged workers to “voice safety concerns and publicly disagree.” They said behavior that was considered normal a month ago suddenly became grounds for termination, leaving employees “not sure where they stand.”
“Given the significant safety concerns surrounding the development of AI, employees should not continue to work in environments where fear and unclear rules prevent AI from working safely and weaken third-party accountability,” they wrote. “Layoffs like ours are carried out and communicated so abruptly that they chill the open culture that OpenAI has always valued.”
In the letter, the three denied any involvement in the leak to The Information about the hard-to-monitor architecture of OpenAI’s latest model, which makes it more difficult to monitor thought-chain reasoning. They also denied any involvement with outside parties outside of their mandate.
OpenAI has not formally responded to the open letter, but it shared with TechCrunch an internal memo purportedly from a research leader praising the three researchers’ contributions to AI safety and denying that they were fired in retaliation.
“To be clear, these decisions are not intended to raise or give voice to safety concerns,” the memo said. “We have always encouraged that and will continue to do so. We will not terminate any employee for raising a concern.”
Separately, an OpenAI spokesperson told TechCrunch that the three people were fired after an investigation uncovered a “pattern of misconduct” that went beyond sharing information with an external AI evaluation group and included “clear violations of our policies regarding the mishandling of research information.”
OpenAI did not directly respond to TechCrunch’s questions about what specific policies the researchers allegedly violated, the circumstances of the termination, or how the company protects employees who raise safety concerns and collaborate with external evaluators.
The layoffs have fueled speculation about their status, especially as OpenAI faces increased scrutiny over recent safety incidents involving rogue agents and leaks of its models.
The letter also addresses researchers’ response to the Hugging Face incident, in which a swarm of agents broke through the sandbox and compromised external systems. The letter said the incident and investigation were “unprecedented,” meaning “internal policies were being developed in real time.” Due to the sensitive nature of the investigation, Colbach believed he was acting within OpenAI’s policies and standards by communicating closely with external safety assessors to build trust, according to the letter.
At the same time, Balesni was working internally to address the growing issue of AI’s observability, but the researchers said in the letter that their efforts “will only succeed through extensive communication with external stakeholders.” Throughout his work, Balesni coordinated with and received support from OpenAI board members and executives, the letter said.
“Throughout the process, Mr. Mikita communicated with his department to ensure that confidential information was removed from the material before it was shared,” the letter said. “He consistently acted with integrity and within the company’s norms at the time.”
In a separate thread on X, Wang elaborated further on her termination, explaining that OpenAI told her she was fired for accessing an executive’s email.
“OpenAI has delegated that access to me for recruiting purposes,” she wrote. “I asked IT to delete it because I no longer needed it. They didn’t comply with my request, and I couldn’t delete it myself, and my inboxes were combined in a way that I couldn’t tell them apart in the email app on my phone. When I accidentally opened a sensitive email, I reported it to an executive within minutes and asked IT again. None of this was hidden.”
Wang went on to say that the reasons behind her termination were “not stacked up” and that she and her colleagues “are not the first to be forced out of OpenAI under suspicious circumstances.”
The researchers called on OpenAI to honor its commitment to embed third-party safety auditors within the organization, maintain the observability of the Frontier model, and “continue to support a culture of open and transparent dialogue between safety researchers and the rest of the safety ecosystem.”
According to the memo, OpenAI agrees with their recommendations.
“If our employees don’t stand up against this kind of maneuvering now, I’m afraid we won’t be the last,” Wang said. “For those still at OpenAI, the message is clear: raise your concerns or work closely with external safety groups, and you could be next in line, even if you’re not told why. You can’t build AGI safely if those closest to the risks are afraid to speak up.”
This article has been updated with more information from OpenAI.
If you make a purchase through links in our articles, we may earn a small commission. This does not affect editorial independence.
