Greg Brockman: OpenAI pulled 25% of production engineers to run Astra on its own systems until no vulnerabilities were left
Anen-o-me · reddit · 2026-09-15
A Reddit post shares a clip of OpenAI president Greg Brockman describing how OpenAI pointed its internal security tool Astra at its own systems, hunting vulnerabilities until it "ran out of vulnerabilities to find."
Brockman says OpenAI took 25% of its production engineers, told them their projects were on hold, and reassigned them to defense — up-leveling the company's security architecture and using AI to assist. The approach illustrates how a frontier lab can systematize security: an AI-driven vulnerability discovery loop paired with a large-scale engineer redeployment, rather than ad-hoc pen testing.
More from Companies & People
- Silicon Valley hearsay: security researchers will soon out-earn AI researchers — eliza_luth · 2026-10-31
- Dev slams Copilot but praises GitHub execs after candid Universe chats — DanWahlin · 2026-10-31
- Sequoia's Maguire pushes back on Ed Zitron's OpenAI jab: "People have no idea what's coming" — shaunmmaguire · 2026-10-31
- Stratechery insight: Palantir's FDE is a product strategy, not just a role — annetgriffin · 2026-10-31
- Google's Web AI Summit 2026 draws 1K+ RSVPs with 24 speakers, 25 edge-AI demos — jason_mayes · 2026-10-31
- Proton CEO cancels forced AI "re-interviews" after exposé; users call for resignation — 404 Media · 2026-10-31