OpenAI reportedly fires three researchers over Signal chats
The three researchers say they never shared confidential information outside the company.
25 Oct 2026 0 reads Compiled with AI
OpenAI fired three of its safety researchers on October 1, citing a breach of trust. The three, Tomek Korbak, Mikita Balesni and Jasmine Wang, had been accused of using Signal to message an outside safety researcher. All three had pushed to keep AI reasoning visible and open to outside scrutiny. OpenAI said the firings were not for raising safety concerns or speaking out.
Investigators reviewed the researchers' personal phones and found that some of their Signal chats with Ryan Greenblatt had disappearing messages turned on. Greenblatt now works at METR. OpenAI said an internal investigation found the three had broken its policies on sensitive information, and that using Signal was not approved. The researchers deny sharing confidential information.
Triggering article about Astra model
The probe began after a September article in The Information reported that OpenAI's Astra model used a new technique that made it more powerful but showed less of its reasoning. OpenAI treated the report as a possible leak of internal design details and set out to trace its source. The episode set off a wider debate about AI reasoning.
The three deny breaking any rules. Balesni said he was dismissed for "speaking too much to third‑party safety organisations". Korbak called disappearing messages a basic security best practice. Wang said she had asked OpenAI to remove her access to an executive's email nearly a year earlier. She said she had opened one sensitive email by mistake last month and reported it at once.
Greenblatt denied the accusations. "Based on this article, OpenAI has made false accusations about me," he wrote on X. He said the three had been told he helped leak architecture details to The Information, which he called false. He also said OpenAI accused him of reading an internal memo with sensitive details. He said he saw only a version without sensitive information, one approved for sharing with him, and believes OpenAI now knows it mixed up the two documents.