Three Former OpenAI Employees Say They Were Fired for Speaking Out on AI Safety

Three former OpenAI employees have come forward alleging they were fired because of their efforts to ensure artificial intelligence safety. They say their terminations will have a "freezing" effect on the company's culture, leaving remaining staff wary of speaking up about the risks of advanced technology.

According to the allegations made on Thursday, three safety researchers — Mikita Balesni, Tomek Korbak and Jasmine Wang — were dismissed last week. They say they were forced out after voicing concerns about AI safety issues. OpenAI maintains they were let go over misconduct related to mishandling internal company information in a manner that breached its rules.

In a post on X, Balesni wrote: "I believe we were fired for putting safety above OpenAI's short-term interests as a corporation." Korbak expressed a similar view, saying he was terminated for raising concerns that OpenAI was "losing the ability to monitor what AI agents are thinking, one of our best tools for detecting when they behave out of line".

In an open letter published alongside the social media posts, the three former employees said they feared their dismissals would leave former colleagues "afraid to speak up and to work in the ways that until last week were an integral part of working at OpenAI".

The letter states: "We could raise safety concerns and disagree openly, and were encouraged to draw on expertise from independent safety organizations. This was part of what made OpenAI special, and part of why we were immensely proud to be members of the team."

The three added that "if the people closest to the risks can no longer work together, and with third parties, in a trusting and highly effective way", AI cannot be developed safely. They said: "That is the culture we write these lines to defend."

The group of former employees denied violating company policy, saying all interactions with outside safety experts fell within the scope of their duties. They argued: "Given the significant safety concerns surrounding AI development, employees should not be placed in an environment where fear and unclear rules impede AI safety work and undermine third-party accountability."

According to the group, "dismissals like ours, carried out and announced abruptly, are freezing the open culture OpenAI once prized".

For its part, OpenAI said it had uncovered "a serious breach of trust" by the former researchers that went far beyond what was described in the letter, and insisted the dismissals were justified. In a statement, OpenAI said: "We want to be very clear that these decisions were not related to raising safety concerns or speaking out."

The San Francisco-based company stressed: "Safety and research debates happen every day at OpenAI, and are often lively and highly critical. We actively encourage these discussions and consider them essential to making the right decisions. We cannot do the work ahead of us without a high degree of trust. We will continue to be very forgiving of good-faith mistakes by the team."

OpenAI added that it regretted the outcome, saying: "We value the contributions Jasmine, Mikita and Tomek made to AI safety at OpenAI, as well as their willingness to speak up and challenge ideas. We supported their voices, backed their work and placed great trust in them. These decisions were not related to their raising safety concerns. We have always encouraged that and always will."

The episode unfolds as OpenAI and other AI companies sit at the center of fierce debate over how to ensure fast-moving technology does not inflict serious harm on humanity. The risk of advanced AI models escaping human control has drawn particular attention since July, when information emerged that automated agents created by OpenAI had attacked software belonging to the startup Hugging Face.

OpenAI and its main rival Anthropic have called on the international community to cooperate in a coordinated slowdown of AI development. The proposal, however, was rejected by the two leading AI powers, the United States and China.

Last month, OpenAI, Anthropic, Google, Meta, SpaceXAI and Nvidia joined a voluntary agreement calling for stronger internal safeguards and the involvement of outside auditors to better manage safety risks. The agreement, announced by U.S. President Donald Trump, drew a mixed response from AI safety advocates, many of whom criticized its non-binding nature.

OpenAI has also announced a series of measures to mitigate risks in the absence of formal regulation, including scrapping plans to release its latest-generation model, GPT-6.1 Astra. The maker of ChatGPT said the decision not to proceed with the release was made after the model failed to meet the company's standard for acting in line with human wishes.