Three safety researchers who were fired by OpenAI last week have published an open letter denying the company's claims that they improperly handled sensitive information outside of established company procedures, and warning that their dismissals signal a chilling effect that will have ripple effects on company culture.

The open letter to OpenAI's safety committee was posted on social media by dismissed employee Mikita Balesni, who previously worked on AI alignment research. The other two employees who signed the letter, Jasmine Wang and Tomek Korbak, both previously worked on AI safety research.

Last week, the three researchers were fired for allegedly leaking confidential company information to third-party AI safety organizations. OpenAI said they "accessed and handled sensitive company information" in violation of company policy. The three individuals said the company only explained the reasons for their dismissal orally, and that none of their actions exceeded the scope of their respective job responsibilities.

In the open letter, the three said they worry that former colleagues may now be "afraid to speak up, and afraid to carry out their work in certain ways." They wrote: "AI is not an ordinary technology, and OpenAI is not an ordinary company. Those of us who work on safety discover risks earlier than anyone else, and we rely on close collaboration with external experts to develop response plans. Being able to carry out this work without fear, and having well-established internal processes to support it, is itself a critical safety mechanism."

They said the dismissals reflect a major shift in OpenAI's corporate culture, noting that the company had previously encouraged employees to "raise safety concerns and openly express dissenting views." They said employees now "don't know where they stand," because behavior that was considered normal just a month ago has suddenly become grounds for termination.

"Given the significant safety risks involved in AI development, employees should not have to work in an environment of fear and unclear rules, which would hinder AI safety work and weaken third-party accountability," they added. "Abruptly terminating contracts like ours, with such rushed enforcement and notification, is killing the open culture that OpenAI once cherished."

Korbak said on social media platform X that he was told the reason for his dismissal was his communication approach with AI safety organization METR. OpenAI had worked with METR to investigate the "Hugging Face" incident. Korbak said he was OpenAI's "primary technical contact" with METR.

Korbak said he believes he was fired because "for several months, I had been raising safety concerns that we are losing the ability to monitor the thinking of AI agents, which is one of our best tools for catching their misbehavior." He also added that he worries his dismissal will become a pretext for OpenAI to stop working closely with METR, which enjoys a strong reputation in the AI safety field.

Balesni agreed with Korbak's account of the reasons for their dismissals, writing on X that he was told he "communicated too much with third-party safety organizations." He said he understood OpenAI to be implying that he leaked OpenAI's intellectual property, but he denied this.

The researchers called on OpenAI to fulfill its public commitment to embed third-party safety auditors within the organization to keep frontier models monitorable, and to "continue supporting an open and transparent culture of dialogue between safety researchers and other members of the safety ecosystem."

Wang said: "Unless employees stand up against this practice now, I fear we won't be the last. The message to everyone still working at OpenAI is clear: if you raise concerns or work closely with external safety organizations, you could be next, and no one will tell you why. If the people who understand the risks best don't dare to speak up, you cannot build artificial general intelligence (AGI) safely."