Ex-Facebook CSO Stamos: humans can't write safe code against AI bug hunters
dyn___ · x · 2026-11-20
Former Facebook CSO Alex Stamos weighs in on the "AI vulnpocalypse" debate:
- Facing superhuman bug-finding machines, humans cannot write safe code without extreme care and planning — a level of rigor almost never applied to real-world critical systems, because software engineers aren't truly "engineers" the way civil engineers are.
- Even careful, experienced engineers struggle to defend C/C++ and other low-level code against modern models.
- CVEs are overrated: huge numbers of organizations are owned by custom vulns in hundreds of billions of lines of non-standard enterprise code, or misconfigurations in heterogeneous systems — which LLMs can already find and write live exploits for.
- The path forward is architectures where a single RCE isn't game over, buying time for defenders.
The quoted post blames security vendors instead, arguing vulnerability announcements are marketing-driven, with fake PoC videos in contrived lab environments creating false impressions of exploitability and exhausting response teams.
More from AGI Musings
- Surya Ganguli's KITP talk 'AI, Math and Physics: the State of Play' draws wide praise — SuryaGanguli · 2026-11-20
- Jeff Sebo: Common sense is making respected thinkers miss the mark on AI — connoraxiotes · 2026-11-20
- ML author Burkov slams AI-written love notes: 'seriously ill' — burkov · 2026-11-20
- Lab security apathy: 'all of SF has read access to lab Slack,' research shared in WeChat groups — AndyMasley · 2026-11-20
- The Economist: Why AI engineers are scared of self-building AI — SammaVaco · 2026-11-20
- Yoav Goldberg: Criticize AI clichés explicitly, don't blame "written by AI" detectors — yoavgo · 2026-11-20