OpenAI terminated three former safety researchers last week after an internal investigation concluded they violated company policies on handling sensitive information. The researchers—Mikita Balesni, Tomek Korbak, and Jasmine Wang—were dismissed for what OpenAI described as a 'significant breach of trust' beyond their stated concerns in a public letter. The company reiterated in a statement on X (formerly Twitter) that their terminations were not related to raising safety concerns or speaking out.
The dismissals have reignited debate over AI safety practices at OpenAI, a company whose technology has faced scrutiny over potential risks. The fired researchers publicly contested their terminations, arguing their firings were motivated by their focus on safety and collaboration with external experts.
OpenAI’s Official Rationale for the Firings
OpenAI stated in a post on X that its internal probe found the researchers had violated clear policies on handling sensitive information. The company emphasized that their decisions were not a response to safety warnings or dissent. In the same statement, OpenAI reaffirmed its commitment to third-party safety assessments, noting it was 'actively finalising contracts' with independent auditors and would announce details in the coming weeks.
The researchers’ dismissal letters, shared publicly, did not explicitly mention policy violations. Instead, they framed their firings as retaliation for prioritizing safety over corporate interests. Balesni wrote on X that he believed the terminations were intended to suppress safety-focused advocacy within the company.
Researchers’ Counterclaims and Public Letter
In a joint letter addressed to OpenAI’s leadership, the former employees argued their dismissals could 'chill the open culture' at the company. They expressed concern that their terminations had made colleagues afraid to voice safety concerns, a practice they claimed was previously encouraged. The letter also urged OpenAI to permanently host independent auditors, a commitment the researchers feared might be abandoned following their firings.
Korbak stated in a separate post that he had raised concerns for months about the company’s ability to monitor AI systems, suggesting this was a key factor in his dismissal. Wang and Balesni echoed these sentiments, asserting that their advocacy for stricter safety measures had been met with punitive action.
Broader Context: AI Safety Debates and Regulatory Scrutiny
The dispute occurs amid growing global attention to AI safety risks. A recent AP-NORC poll found that nearly two-thirds of Americans believe AI is developing too quickly. OpenAI has faced bipartisan pressure in the U.S. Senate over potential liability for AI-related security breaches, as well as a copyright lawsuit from USA Today. The company has also warned about the dangers of unchecked AI advancement, positioning itself as a leader in advocating for responsible development.
OpenAI’s leadership has repeatedly emphasized the importance of 'monitorability' of advanced AI models, a principle the fired researchers’ letter also supported. The company confirmed it continues to invest in this area, despite the internal conflict.
Industry Reactions and Unanswered Questions
The terminations have drawn mixed reactions from AI safety advocates. Some industry observers argue the firings highlight tensions between corporate priorities and ethical oversight in AI development. Others suggest the policy violations cited by OpenAI may reflect broader challenges in balancing transparency with proprietary protections in the tech sector.
OpenAI has not provided further details about the nature of the 'breach of trust' or the specific policies violated by the researchers. The company has reiterated its stance that safety concerns were not a factor in the dismissals, while the researchers maintain their terminations were punitive.
The dispute remains unresolved, with both sides presenting competing narratives about the motivations behind the firings and their implications for OpenAI’s culture and future policies.