Matthew Green asks: who pre-positions open models for when frontier AI goes dark?
matthew_d_green · x · 2026-10-19
Cryptographer Matthew Green raises an under-discussed resilience question: what happens in a major incident when all frontier models become unreachable at once? His suggested answer is broader distribution—pre-positioning open-weight models so emergency responders can actually use them. He doubts anyone is seriously doing this today, sparking discussion about single points of failure in AI infrastructure and open models as a disaster-recovery resource.
Related event: Cryptographer Asks What Happens If All Frontier Models Go Down(2 posts)→
More from AGI Musings
- Economists clash on AI productivity: why labor productivity beats noisy TFP right now — danielrock · 2026-10-19
- Ethan Mollick: don't give agents your bank account, but secure design beats bans — emollick · 2026-10-19
- "Neither the best nor worst AI outcome appeals to me": a melancholy middle view — birchlse · 2026-10-19
- The web is walling off AI agents: Amazon, Yelp, Cloudflare all block bots that skip ads — alex_verem · 2026-10-19
- OpenMined founder: information-aggregating tech centralizes before federating, and AGI may follow — iamtrask · 2026-10-19
- Anthropic co-founder Jack Clark: convincing evidence the Singularity has begun — kimmonismus · 2026-10-19