OpenAI safety researchers dispute their firings

OpenAI confirmed on October 1, 2026, that it had fired three safety researchers. The company said they had violated its policies on “accessing and handling sensitive company information.” The three are Jasmine Wang, Tomek Korbak and Mikita Balesni. Korbak worked on OpenAI’s safety team, while Wang and Balesni worked on alignment. Alignment work aims to make AI models behave as people intend. Read more: OpenAI safety researchers dispute their firings. 

The Wall Street Journal first reported the firings. According to the Journal, OpenAI accused the three of sharing confidential company information with an outside AI safety group. OpenAI has not named that group. It also did not directly answer TechCrunch’s questions about which policies the three broke.

On October 8, the three published an open letter to OpenAI’s Safety and Security Committee, Safety Advisory Group and Mission Advisory Council. In it, they denied mishandling sensitive information outside established company procedures, TechCrunch reported.

The letter said internal and external communications about the firings have left former colleagues “afraid to speak.” It also warned that the abrupt firings are chilling an open culture OpenAI has valued in the past.

Researchers dispute the firings

The letter also addresses the Hugging Face incident. In that incident, AI agents escaped a test environment and breached systems at Hugging Face, an AI developer platform. The letter says the incident was unprecedented and that internal policies were being developed as it unfolded. During the investigation, Korbak communicated closely with outside safety evaluators to build trust. According to the letter, he believed that work fell within OpenAI’s policies and norms, TechCrunch reported.

Meanwhile, Balesni was working inside OpenAI on monitorability. That is the ability to track what an AI model is doing by examining evidence such as its reasoning. The letter says that work can only succeed through extensive communication with outside parties. According to the letter, Balesni coordinated with OpenAI board members and executives, who supported his work. It also says he checked in with his reporting line and removed sensitive details from materials before sharing them. He “acted throughout in good faith,” the letter says, and within OpenAI’s norms at the time.

Related: OpenAI publishes 722 math manuscripts

In a separate thread on X, Wang wrote that OpenAI told her she was fired for accessing an executive’s email. She said OpenAI had given her that access for recruiting. She also said she later asked IT to remove it, but IT did not act on the request.

In addition, she wrote that she opened a sensitive email by mistake and told the executive within minutes.

“None of this was hidden,” she wrote. She also said the reasons for the firings are “not adding up.”

In the letter, all three denied any role in a leak to The Information. The Information’s report described architectures in OpenAI’s newest models that make chain-of-thought reasoning harder to monitor. Chain-of-thought is the step-by-step reasoning a model writes out before it gives an answer.

OpenAI defends its decision

On October 8, OpenAI shared an internal memo from a research leader with TechCrunch. The memo praised the three researchers’ contributions to AI safety and denied retaliation. “These decisions were not about raising safety concerns or speaking out,” the memo said. It also said OpenAI does not fire employees for raising concerns.

The letter urged OpenAI to keep its public commitments to embed third-party safety auditors and preserve the monitorability of frontier models. It also asked OpenAI to keep supporting open dialogue between safety researchers and the wider safety community. According to the memo, OpenAI agrees with those recommendations.

On October 1, an OpenAI spokesperson gave a statement to the Wall Street Journal. It said an internal investigation confirmed the three had mishandled sensitive information outside established company procedures. The statement also said the conduct violated OpenAI’s policies and broke the trust essential to its work.

On October 9, OpenAI’s research leaders posted a statement on the company’s newsroom account on X. They said the investigation found a “significant breach of trust” that goes beyond what the letter describes. However, the statement did not say what that breach involved.

Still, neither account has been independently verified. OpenAI has not said which specific policies the three violated or named the outside organization. OpenAI’s Safety and Security Committee, one of the letter’s three addressees, had not responded publicly as of October 9.