Fired OpenAI Safety Researchers Publish Letter Warning of Chilling Effect and Demanding Third-Party Audits

On October 9, three former OpenAI researchers—Mikita Balesni, Tomek Korbak, and Jasmine Wang—who were abruptly fired last week, published an open letter today. The letter expresses concern that this move may create a "chilling effect" on OpenAI's corporate culture, discouraging other employees from speaking out publicly.

In the open letter titled "OpenAI Cannot Ensure AI Safety Alone," the three stated that OpenAI is "not just any company." They emphasized that the ability for employees to confidently collaborate with external organizations to address safety issues has been a vital component of OpenAI's work culture.

The open letter was posted on social media by Balesni, who previously worked on AI alignment research, and addressed to OpenAI's various safety committees. The other two signatories, Korbak and Wang, also previously conducted AI safety research and were similarly dismissed.

According to Business Insider, an OpenAI spokesperson responded by stating that after an internal investigation, the company determined that the three individuals violated policies regarding access to and handling of sensitive information.

The three researchers separately stated on social media that they were only verbally informed of the reasons for their dismissal, and that their prior actions fell within the scope of their respective job responsibilities.

• Jasmine Wang was accused of accessing an executive's email. In response, she said: "The reason given for my termination simply does not hold up."
• Balesni posted that the company told him he was no longer trusted because he had "too much communication with third-party safety organizations." He added: "I never leaked the company's intellectual property externally. My work has always been coordinated with my direct supervisor, research leadership, and the board."
• Korbak also posted that the company verbally told him his dismissal was due to "the way he communicated with METR." He clarified: "Let me be clear, communicating with METR was literally part of my job."

IT Home Note: METR is a non-profit research organization responsible for evaluating AI models and their associated safety risks.

、

In their open letter, the three expressed concern that other employees might now be afraid to speak up or conduct their work as they did before, activities that were integral to OpenAI's operations until last week. "Abruptly firing employees like us, with such rushed notification, delivers a chilling blow to the open culture OpenAI once prided itself on."

The trio also proposed three recommendations to OpenAI: allow third-party safety auditors to embed within the company, ensure that frontier AI models remain subject to effective monitoring, and support safety researchers in maintaining "open and transparent communication" with other institutions and individuals in the AI safety field.

A head of an OpenAI research department strongly supported these three proposals in an internal memo sent to employees. "We are deeply grateful for their contributions to AI safety work, and for their willingness to speak frankly and challenge prevailing views. We value their input, support their work, and placed great trust in them. I must make it clear that the decision to terminate them was unrelated to them raising safety concerns or speaking publicly. We have always encouraged employees to voice their opinions and will continue to do so. We do not fire employees for raising concerns."

Korbak further wrote on social media that he worried OpenAI might use the dismissal of the three as a "pretext" to reduce collaboration with METR.