Insight from Beating AI News: OpenAI fired three security researchers at the end of September for violating sensitive information management policies. The Wall Street Journal disclosed further details on October 22: the company suspected the three had leaked confidential model architecture to external security expert Ryan Greenblatt, demanding they unlock their private phones. Investigators found that some of them had enabled message auto-deletion in their Signal chats with Greenblatt. A week later, all three were summoned to a meeting and subsequently terminated.

Greenblatt had previously participated in OpenAI's external security audits, while one of the fired researchers was responsible for collaborating with external security organizations. All three denied leaking information, believing their termination might be linked to their advocacy for AI safety oversight. One of them claimed he was fired due to accidentally opening an executive email he had access to.

OpenAI denies retaliation, emphasizing that the firings stemmed from internal investigations uncovering policy violations, not from employees raising safety concerns. However, the company has not released full details of the violations, and the three individuals stated they never received written explanations.