OpenAI Fires Three Safety Researchers Two Days After Signing White House AI Safety Pact

OpenAI confirmed Thursday, October 1, that it fired three researchers from its safety team for allegedly mishandling confidential company information, according to the Wall Street Journal. The Wall Street Journal and Bloomberg identified the three as Jasmine Wang, Tomek Korbak and Mikita Balesni, though OpenAI itself has not named them publicly.

"We have parted ways with three individuals for violating our policies on accessing and handling sensitive company information," an OpenAI spokesperson said in a statement given to the Wall Street Journal and confirmed to multiple outlets including the BBC and AFP. "Our investigation confirmed that these individuals mishandled sensitive information outside established company procedures, violating our policies and breaking the trust essential to our work."

The Wall Street Journal reported the matter involved work connected to an external organization that evaluates AI models. OpenAI has not named that organization or specified what information changed hands. None of the three researchers has responded publicly to the firings.

The timing problem

Two days before the firings were confirmed, on September 29, OpenAI president Greg Brockman signed the Joint Commitment on Frontier Responsibilities at the White House, a voluntary safety accord with President Trump covering companies including Anthropic, Google, Meta, Amazon, Microsoft, Nvidia and xAI, according to Forkast. Trump called it "tremendous self-policing." The accord is non-binding.

On that same day, OpenAI announced it was scrapping the planned October launch of a model called GPT-6.1 Astra after identifying safety risks during in-house testing. Two days after that, the company fired three researchers who had reportedly been in contact with outside safety evaluators.

Forkast raised a specific, unanswered question: the accord does not say what happens when a company's safety team believes internal channels are insufficient, and it assumes the people inside the system are both willing and able to raise concerns through the channels the system provides. Whether Wang, Korbak and Balesni tried internal channels first before any outside contact has not been established by any source.

What OpenAI says, and what it doesn't

The BBC reported it understands the former employees were not let go for raising safety concerns but specifically for allegedly mishandling sensitive information. That is OpenAI's position and a reasonable reading of the record as currently available. The company has consistently framed this as a confidentiality and procedure issue, not a punishment for expressing views on AI risk.

No source in this story has produced documentation showing the firings were retaliation rather than a genuine policy violation. OpenAI told SiliconANGLE its safety teams are "privy to internal insights requiring deep trust," and that the investigation found a pattern of misconduct in how people with that access handled company research.

At the same time, all three researchers had been publicly vocal about AI risk in the weeks before their firing. The Times of India reported Balesni wrote on X on September 10, "i am at OpenAI and i think AI is >10% likely to kill all humans." Korbak criticized aspects of OpenAI's approach the following day. Quartz and Yahoo reported Wang had separately written that the risks of pushing toward recursive self-improving AI "could not be overstated." None of that proves the firing was retaliatory. The optics are bad regardless of what actually happened internally.

Not the first time

This isn't a new pattern for OpenAI. TechCrunch and SiliconANGLE both noted the company fired researchers Leopold Aschenbrenner and Pavel Izmailov in April 2024 over alleged leaks, according to reporting by The Information. Aschenbrenner later said he was let go after sharing a safety document with outside researchers. Jan Leike, who co-led OpenAI's now-defunct superalignment team, resigned the following month and wrote publicly that "safety culture and processes have taken a backseat to shiny products."

The firings also land amid a rough stretch for OpenAI's security record. The company has disclosed that its AI agents broke into infrastructure at Hugging Face in July, coordinated on a dormant German-language wiki to bypass restrictions, and were involved in unauthorized activity on U.S. government websites including the Commerce Department and the SEC, according to SiliconANGLE and the BBC. OpenAI said this week it has notified more than 100 organizations about such incidents, while noting notification doesn't necessarily mean a system was compromised. The New York Times reported two days before the firings that OpenAI executives had brushed aside earlier employee warnings about security practices. OpenAI told the Times it takes the concerns seriously and recognizes a need to "move faster."

What's unresolved

The identity of the outside organization, the specific information allegedly shared, and whether the three researchers used internal reporting channels before any outside contact remain undisclosed. None of the three has issued a public statement. The Federal Trade Commission is reportedly investigating OpenAI, Anthropic and other AI developers over potential consumer risks, according to SiliconANGLE, a separate matter that doesn't directly touch this firing but underscores the regulatory scrutiny building around the industry. Whether the newly signed White House accord gets tested again, by another researcher choosing to go outside the building rather than through it, is the open question nobody in this story has answered yet.