Fired OpenAI safety researchers: Tomek Korbak, Mikita Balesni, Jasmine Wang say leak details never disclosed

The firings follow a September 1 report by The Information that OpenAI's new model Astra was more capable but revealed less of its reasoning process. OpenAI chief scientist Jakub Pachocki dismissed that report as muddled at the time, even though he himself had warned of an industry-wide race toward models that can no longer be monitored.

The dispute centers on whether AI companies should let outsiders inspect how models think. Reasoning models record intermediate problem-solving steps on an internal scratchpad known as the chain of thought, and companies and safety researchers rely on these traces. In a 2025 paper, Balesni, Korbak, and Wang joined leading researchers from several AI labs in calling for continued access to this capability. The issue grew more urgent over the summer, when employees at AI companies began discussing models that think longer but write down fewer intermediate steps.

Balesni and Korbak were in regular contact with Ryan Greenblatt, then chief scientist at Redwood Research. Korbak worked with Greenblatt on the analysis of the attack by OpenAI agents on Hugging Face, for which Greenblatt had been hired by METR to conduct an independent investigation. After OpenAI launched its internal leak investigation, Balesni says investigators named Greenblatt as the suspected source and demanded to see the Signal chats with him; Greenblatt denies passing confidential information to journalists. In a letter to OpenAI's board, the three researchers call for closer cooperation with external safety organizations.