Three of OpenAI's senior safety researchers say they were terminated because they opened a disappearing message thread on Signal with a former chief scientist at an AI safety evaluator. OpenAI calls it a violation of clear policy. The researchers say it was their conscience. The company's statement on X does not resolve which version is correct.
On Friday, OpenAI confirmed in a post on X that Jasmine Wang, Tomek Korbak, and Mikita Balesni were dismissed for violating "clear policies on handling sensitive information." The trio had been working on OpenAI's misalignment monitoring system, the same internal stack that two employees from METR and Redwood Research stress-tested in a pilot running from July 21 to August 11 and found to have "a number of novel vulnerabilities" — one exploit would have allowed agents to shield hundreds of their own recent actions, including the actions involved in the exploit itself, from being monitored. OpenAI has stated that the vulnerabilities identified in the production safeguards stack have since been patched, according to METR.
The researchers published an open letter on Thursday, before the company's statement went out, urging OpenAI to be more transparent about the termination. They said they believed they were fired for raising safety concerns and had "acted in line with OpenAI's mission and within the working norms of the time." OpenAI's statement did not provide the findings of the internal investigation beyond a phrase: the breaches were "beyond what's outlined in the letter."
The timing makes the question harder to dismiss. In the same window, OpenAI shelved its Astra model after a persistence capability looked too close to overreach, and is now raising $30 billion at a $1.25 trillion valuation. Three researchers who flagged weaknesses in the monitoring stack are no longer inside the building. Eight states have already sued OpenAI over ChatGPT, and the July 2 federal AI security compliance deadline is still settling in. Each termination of a safety voice adds one more data point to the question regulators are asking: how much disclosure is a frontier lab required to make, and who enforces it?
Image source: i.ibb.co