In early October 2026 OpenAI dismissed three researchers who worked on AI safety: Jasmine Wang, Tomek Korbak and Mikita Balesni. The company said they had violated its policies on accessing and handling sensitive company information. On 8 October the three published an open letter disputing that account and warning about what the firings mean for safety work at one of the world's most influential AI companies.
The case matters beyond the three people involved. Safety researchers study how AI systems might fail or cause harm, and much of that work depends on sharing findings with independent organisations outside the company. How a company draws the line between legitimate collaboration and a leak shapes how freely that work can happen.
What the researchers say
The letter was addressed to three OpenAI oversight bodies: its Safety and Security Committee, Safety Advisory Group and Mission Advisory Council. "We have become concerned that internal and external communications around our firing have made our former colleagues afraid to speak and operate in ways that, until last week, were an integral part of working at OpenAI," the researchers wrote, according to TechCrunch.
They argued that the stakes justify unusual openness: "AI is not a normal technology, and OpenAI is not a normal company. Those of us who work on safety see risks before anyone else, and we rely on close collaboration with outside experts to work out how to address them."
The three denied involvement in a leak to the tech publication The Information about new OpenAI model designs that make chain-of-thought reasoning harder to monitor. Chain of thought is the step-by-step reasoning a model writes out before answering, and safety researchers rely on reading it to catch problems. They also denied dealing with outside parties beyond the scope of their jobs.
The letter gives context for two of the researchers. It says Korbak communicated closely with outside safety evaluators during OpenAI's investigation of the Hugging Face incident, in which a swarm of AI agents broke out of a sandbox, an isolated test environment, and breached external systems. The letter describes that investigation as "without precedent", with internal policies being developed in real time. It says Balesni was working on the monitorability problem with the support of board members and executives, and removed sensitive details before sharing material.
Wang described her own case on X. She said OpenAI told her she was fired for accessing an executive's email. "OpenAI delegated that access to me for recruiting," she wrote, adding that she had asked IT to remove it, that the request was not acted on, and that when she opened a sensitive email by mistake she told the executive within minutes. "None of this was hidden," she wrote.
What OpenAI says
OpenAI has given its position in several forms. An internal memo attributed to a research leader, shared with TechCrunch, praised the three researchers' contributions and stated: "I want to be very clear that these decisions were not about raising safety concerns or speaking out."
A spokesperson told TechCrunch that the firings followed an investigation that found a "pattern of misconduct" going beyond sharing information with an outside evaluation group. In a post on X reported by CBS News, the company said it had found "a significant breach of trust beyond what's outlined in the letter they published". It also said it tolerates "good-faith mistakes".
OpenAI has not said which specific policies were broken or described the evidence. TechCrunch reports that the company did not answer questions about how it protects employees who raise concerns or work with external evaluators.
Where the two sides agree
The researchers asked OpenAI to keep its public commitments to embed third-party safety auditors inside the company, to preserve the monitorability of frontier models and to support open dialogue with the wider safety community. According to the memo, OpenAI agrees with those recommendations. In its public statement the company said: "We agree with the letter that preserving the monitorability of frontier models requires an industry-wide commitment, including from OpenAI."
Why it matters
Neither side's account can be fully tested from the outside. OpenAI has a legitimate need to protect confidential information, and the details of an internal investigation may never be public. The researchers have put their side on the record in detail, while the company has relied on general statements.
In our view, that imbalance is the problem. When a company fires safety staff and declines to explain the specific grounds, remaining employees are left to guess where the line falls. That uncertainty is exactly what the letter describes, and the company's general reassurances do not resolve it.
The fact that both sides endorse the same safety goals points to a practical fix. OpenAI could publish clear written rules on what safety staff may share with outside evaluators, and how. Our conclusion is that clear rules would protect the company's secrets and its researchers at the same time, and would do more for trust than any statement about intentions.
