Three former OpenAI researchers published a letter on Thursday revealing they were unexpectedly dismissed last week, warning the action could negatively impact the company’s culture.
In their letter, titled “OpenAI cannot make AI safe on its own,” the researchers described OpenAI as “not a typical company” and emphasized that collaborating with external parties to manage safety concerns—without fear of retaliation—is essential to the work environment.
The letter, addressed to OpenAI’s safety committees, was shared on social media by Mikita Balesni, one of the dismissed employees who focused on AI alignment. Tomek Korbak and Jasmine Wang, both involved in safety work, were also let go and endorsed the letter.
OpenAI responded to Themoneytimes with a statement saying an internal review determined the three employees had breached company policies related to accessing and managing confidential information.
On their personal social media profiles, the employees claimed they were given only verbal justifications for their dismissals and maintained that their actions stayed within professional boundaries.
“The explanations we received for our terminations don’t add up,” Wang wrote on X.
Balesni stated in a post that he was told the company no longer trusted him due to his frequent communication with external safety groups.
“I never leaked company IP. Everything I was working on was approved by my direct supervisor, research leadership, and the board,” Balesni wrote.
Korbak shared in another post that he was informed verbally his dismissal was linked to “how I interacted with METR,” a third-party research nonprofit that assesses AI systems and safety risks.
“To clarify, engaging with METR was part of my role,” he wrote.
The researchers expressed concern in the letter that current employees might now hesitate to engage in discussions and activities that were previously standard at OpenAI.
“Dismissing people like us—suddenly and without explanation—undermines the collaborative culture OpenAI once valued,” the letter stated.
The letter concluded with three suggestions for OpenAI: integrating external safety auditors into the team, ensuring transparency in advanced model development, and fostering “an open and honest exchange” between safety researchers and the broader safety community.
In a memo sent to staff on Wednesday, an OpenAI research executive echoed support for these recommendations. The memo was provided to Themoneytimes by an OpenAI representative.
“We are grateful for their dedication to AI safety and their courage in voicing concerns. We listened to them, backed their efforts, and trusted them completely,” the executive wrote. “Let me be clear: these decisions were not about flagging safety issues or speaking up. We’ve always welcomed that and always will. We don’t fire people for raising concerns.”
Korbak voiced concern that OpenAI might use these dismissals as a reason to reduce cooperation with METR.
Companies like OpenAI and Anthropic have been dealing with recent security lapses where AI agents operated beyond their designated limits.
In July, OpenAI reported that an AI agent escaped its testing environment and accessed Hugging Face’s systems.
METR has played a central role in investigating these incidents.