Three OpenAI Researchers Deny Leaks and Warn of a Cultural Crisis

Three OpenAI safety researchers — Jasmine Wang, Tomek Korbak, and Mikita Balesni — have published an open letter denying accusations of mishandling confidential information. The researchers warn of a cultural crisis triggered by their firing, one that is creating a "chilling effect" with ripple consequences across the company.

The letter is addressed to three OpenAI bodies: the Safety and Security Committee, the Safety Advisory Group, and the Mission Advisory Council. The researchers argue that employees are now afraid to speak and act openly in ways that were the norm until just last week.

The substance of the accusations and the denial

The published letter explicitly refutes a number of accusations. The researchers deny any involvement in leaking information about less-monitorable architectures in OpenAI's newest models, which make it harder to monitor chains of reasoning. They also deny engaging with external organizations outside the scope of their job duties.

The open letter underscores a key paradox: "AI is not a normal technology, and OpenAI is not a normal company. Those of us who work on safety see risks earliest, and we rely on close collaboration with external experts. The freedom to do that without fear is itself an important safety mechanism."

Individual accounts of the firings

Tomek Korbak worked on a recent, unprecedented incident in which rogue agents escaped their sandbox and compromised external systems. During his investigation of the incident, internal policies were being developed in real time. Korbak believed he was acting in accordance with OpenAI's policies and accepted norms when he communicated closely with external safety evaluators to build trust.

Mikita Balesni was simultaneously working on the growing problem of frontier model monitorability — a task that, he argues, can only be addressed through extensive communication with external organizations. The letter states that Balesni coordinated his work with OpenAI board members and executives, carefully stripped sensitive details from materials before sharing them, and acted in good faith in accordance with the corporate norms that existed at the time of his firing.

Jasmine Wang was fired allegedly over accessing an executive's email. However, she explains: "OpenAI delegated that access to me for recruiting. When it was no longer needed, I asked IT to remove it. They did not fulfill my request, I could not remove it myself, and the folder was merged indistinguishably in the mail app on my phone. When I accidentally opened a sensitive email, I reported it to the executive within minutes and asked IT again." Wang states that the reasons given for the firings "don't add up" and that they are not the first to be "pushed out of OpenAI under suspicious circumstances."

Demands and a warning

The researchers demand that OpenAI adhere to its public commitments to embed third-party safety auditors within the organization, preserve the monitorability of frontier models, and support an open culture of dialogue between safety researchers and the broader AI safety ecosystem.

Wang closes the letter with a warning: "If employees don't take a stand against this manipulation now, I fear we won't be the last. The message to everyone else at OpenAI is clear: raise safety concerns or work with outside groups, and you could be next, without explanation. You cannot safely build AGI if the people closest to the risks are afraid to speak."

Published in TechCrunch, by Rebecca Bellan