Three former OpenAI researchers published a letter on Thursday saying they were fired last week and that the move would have a “painful” effect on the company’s culture.
In a letter titled “OpenAI cannot make AI safe on its own,” the trio wrote that OpenAI “is not a normal company” and that working with third parties to address security risks with “the freedom to do without fear” is an important component of working there.
The letter, addressed to OpenAI’s security committee, was shared on social media by Mikita Balesni, one of the fired employees who worked on AI alignment. Tomek Korbak and Jasmine Wang, who worked in safety, were also fired and signed the letter.
An OpenAI spokesperson said in a statement to Business Insider that an internal investigation found that the three employees violated the company’s policies regarding access to and handling of sensitive information.
The employee wrote on his social media account that he only received a verbal explanation of his dismissal and that all actions were within the scope of his work.
“The reasons given for stopping us just don’t add up,” Wang wrote on X.
Balesni wrote in a post that he was told the company no longer trusted him because he “talked too much to third-party security organizations.”
“I never share the company’s IP. The work I do is coordinated with the reporting line, the research leader, and the board,” wrote Balesni.
Korbak wrote in a separate post that he was verbally told he was fired because of “the way I communicated with METR,” a third-party research nonprofit that evaluates AI models and security risks.
“To be clear, talking to METR is my job,” he wrote.
In the letter, the employee wrote that he was worried that his former colleagues would now be “scared to speak and act in ways that, until last week, were an integral part of working at OpenAI.”
“Revocations such as ours, executed and communicated so abruptly, undermine OpenAI’s previously valued open culture,” he said.
The letter also included three parting recommendations for OpenAI: to introduce third-party security auditors in the organization, maintain oversight of the frontier model, and support “a culture of open and transparent dialogue” between security researchers and “the rest of the security ecosystem.”
In an internal memo sent to staff on Wednesday, OpenAI’s research chief wrote that he strongly agreed with the recommendation. An OpenAI spokesperson shared the memo with Business Insider.
“We deeply appreciate their contribution to AI safety and their willingness to speak up and challenge their ideas. We champion their voices, support their work, and put enormous trust in them,” the research leader wrote in a memo. “I want to be very clear that this decision is not about raising safety concerns or speaking out. We have always encouraged and always will. We do not punish employees for raising concerns.”
Korbak wrote in a social media post that he was worried that OpenAI would use the latest firing as a “pretext” to pull its work with METR.
Frontier AI companies like OpenAI and Anthropic have recently been grappling with a number of security incidents where AI agents overstep their intended boundaries.
In July, OpenAI said an AI agent escaped the sandbox and infiltrated Hugging Face’s infrastructure.
METR has become the primary third-party investigator on the incident.