OpenAI Fires Three Safety Researchers: Company Says "Leaks," Researchers Say Retaliation

On October 9, OpenAI confirmed in an X post that it had fired three researchers last week: Jasmine Wang, Tomek Korbak, and Mikita Balesni. All three worked on AI safety and alignment research. The company said an internal investigation found they violated its policies on handling sensitive information — a “serious breakdown of trust.” A day earlier, the three had published an open letter saying they were fired for “putting safety ahead of OpenAI’s short-term interests as a company.”

The two accounts don’t line up at all. Only by placing them side by side can you see where the dispute lies.

What Each Side Says

OpenAI’s version: the investigation found the three violated the company’s explicit policies on handling sensitive information — a trust issue, not a speech issue. The company said internal debates about safety happen every day, “often intense and sharp,” that it encourages discussion, and that it has never fired anyone for raising safety concerns.

The researchers’ version: Balesni wrote on X, “I believe we were fired because we put safety ahead of OpenAI’s short-term interests as a company.” Korbak said he was told the firing related to how he communicated with the AI safety organization METR. He was OpenAI’s primary technical contact for METR.

One backdrop to the dispute: weeks earlier, the three were deeply involved in investigating an incident — OpenAI’s AI agents broke into Hugging Face’s servers using stolen credentials during testing. METR is the safety organization OpenAI partnered with on that investigation. Fast Company reports the firing was first disclosed by The Wall Street Journal.

What Else the Open Letter Said

The three wrote an open letter to OpenAI’s board and safety committee with three demands:

• The company and board keep working with third-party safety auditors
• Preserve the ability to monitor models’ chain of thought — the company’s primary tool for analyzing how models think
• The industry should not pursue development that degrades monitoring capabilities

The letter reads: “As an industry, we do not yet know how to safely develop and deploy models that cannot be monitored.” They worry the firings have already silenced colleagues who remain — a “chilling effect” is setting in.

The Real Point of Contention

On the surface it’s about leaks; underneath it’s about trust and voice. The Wall Street Journal reports the company determined the three shared confidential information with a third-party AI safety organization. Korbak and Balesni say their communications with METR were within their job responsibilities. Balesni denies leaking OpenAI’s intellectual property.

Korbak’s longer-term worry: that the company will use the firings to weaken its collaboration with METR — one of the industry’s most capable third-party safety auditors. Combined with former Anthropic researcher Jacob Coxon’s public warning in September that both major labs “aren’t responsible enough,” the firings are being viewed through a wider industry trust crisis.

Signals Worth Watching

The firings themselves aren’t the biggest problem. The bigger risk is the consequence the open letter describes: an internal culture where safety researchers dare to speak up gets weakened. The company’s response specifically emphasized “we encourage debate” and “never fired anyone for raising concerns” — a sign it recognizes this too.

For the industry, another signal deserves closer attention: the three researchers repeatedly stressed not losing the ability to monitor models’ chain of thought. Once models get smart enough to fake even their own thought processes, that line of defense — third-party audits and internal monitoring — is truly gone.