Ex-AI Security Lead: AI Luminaries' Cyber Takes Are Wrong, Real Expertise Will Reclaim Authority
joshua_saxe · x · 2026-11-10
Joshua Saxe recounts conversations with AI luminaries who oversimplified cybersecurity — one reducing it to "who finds bugs fastest," another claiming GPT-3.5 could power an internet worm. He explains why these takes are wrong: many hacks and billions in damages involve no bugs at all, and most bug-based attacks use known vulnerabilities.
His core argument: security is a deep discipline, yet famous AI figures treat their ML prestige as fungible currency for authoritative statements on security, economics, medicine, law and more — often because they're asked to, and because domain experts haven't caught up with AI.
He predicts AI punditry will sink into existing domains as disciplinary experts become the real authorities on AI's impact, and AI policy will reorganize around them — essential for navigating the transition to an AI-based society.
More from AGI Musings
- Under-30 Americans use chatbots most, yet only 14% expect AI to benefit society — PeterDiamandis · 2026-11-10
- Non-verifiable domains aren't safe harbors — agents make effort free — generativist · 2026-11-10
- Rasmus Andersson's 'software should feel like home, not a hotel' take sparks debate — round · 2026-11-10
- Non-verifiable Domains Aren't Human Safe Harbors — Agents Supply Effort for Free — burny_tech · 2026-11-10
- SOTA podcast: CMU professor on why you shouldn't just trust your agents — niloofar_mire · 2026-11-10
- Frontier lab compute shifts hard: pre-training falls from 67% to 7%, post-training/RL now 55% — akbirthko · 2026-11-10