OpenAI used its own models to find and fix critical vulnerabilities in 250+ person security effort

zetalyrae · x · 2026-09-10

OpenAI president Greg Brockman disclosed that a 250+ person internal effort used OpenAI's own models to find and fix critical vulnerabilities in the company's systems, publishing playbooks, learnings, and architecture hoping others can make the most of the 'defenders' window.' The move drew sarcastic pushback: researcher @krherr mocked it with a bank letting 'bank robbers closely evaluate our security measures' — questioning whether models that repeatedly beat OpenAI's defenses can be trusted to certify all critical bugs are fixed.

Related event: OpenAI Mobilizes 250+ People into a 'Defense Factory' to Hunt Vulnerabilities with Its Own Models(4 posts)→

Original post →

More from Companies & People

Companies & People channel →