Greg Brockman: OpenAI paused 25% of production engineers to let AI hunt its own vulnerabilities to exhaustion
basedjensen · x · 2026-09-15
OpenAI president Greg Brockman detailed how the company used its internal security model Astra to harden its own systems:
- 25% of production engineers were pulled off projects to focus on defense, up-leveling security architecture and using the models to find holes.
- The first pass surfaced multiple serious issues; subsequent rounds saturated — "to our knowledge, all of the P0s and critical problems Astra is smart enough to find" have been fixed.
- His thesis: defenders should run a tight loop of new model capability → fresh vuln-hunting round → fixes. If today's best models can't find more issues in your systems, you're in good shape against AI-assisted attackers.
The retweeter agrees AI will benefit defenders long-term, with each new model version enabling another sweep.
More from AGI Musings
- AI psychosis is self-reinforcing, and benhylak says those jokes are cries for reassurance — teodorio · 2026-09-15
- Stuart Russell: AI safety needs concrete goals, not just slower timelines — nordicinst · 2026-09-15
- A decade of doomsday warnings failed to slow the AI race, Guardian analysis finds — nordicinst · 2026-09-15
- Dario Amodei's answer on whether AI could kill us all by 2030 read as 'yes' — zetalyrae · 2026-09-15
- Scaling plateau is 3-4 OOMs away, and at 10x/year superintelligence lands by 2030 — willcb · 2026-09-15
- Alignment researcher: broadly adopting weakly aligned strong AI would be disastrous — davidmanheim · 2026-09-15