Anthropic's 'inevitable' framing on AI cyber risks draws fire over weak distillation protections
DKokotajlo · x · 2026-10-24
Nathan Calvin responds to Anthropic's head of security on AI cybersecurity risks with three critiques:
- Many current AI cyber risks would likely have happened regardless, and some Anthropic warnings look prescient — but Anthropic's weak protections against distillation despite being warned accelerated capabilities in Chinese open-weight models now used in attacks
- The "it was inevitable" logic doesn't hold for RSI risks, where Anthropic is genuinely intensifying the race and counterfactually contributing to risk
- As a comms strategy, "not our fault" is unlikely to resonate — the public will blame AI companies for AI-related harms
A substantive debate over frontier labs' responsibility boundaries for security risks.
More from Companies & People
- LatePost Details ByteDance Intern Sabotage Case and the AI Talent War Era — dotey · 2026-10-24
- Amazon Research Awards funds multilingual AI evaluation for low-resource languages, due Nov 4 — Fowe · 2026-10-24
- Hustle Fund GP: our fund now has nearly as many AI agents as full-time staff — dunkhippo33 · 2026-10-24
- 9 out of 10 engineers can't state your strategy — and that team costs $1.4M a year — aakashgupta · 2026-10-24
- 9 out of 10 engineers can't state your strategy: 19 rules of product strategy — aakashgupta · 2026-10-24
- Stratechery: Apple×Meta Edge AI Partnership, Netflix's 40% Stock Slump, Stripe Sponsorship — Stratechery · 2026-10-24