OpenAI used its own models to find and fix critical vulns in a 250+ person defense effort

gdb · x · 2026-09-10

OpenAI says a 250+ person internal effort used its cyber models to find and fix critical vulnerabilities across hundreds of its own systems—bugs it might never have caught otherwise.

The company is publishing the architecture, learnings, and a practical playbook so others can build a "Defense Factory": a continuous loop where AI agents discover vulnerabilities, validate them, and verify that fixes work, exploiting the defenders' window.

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 →