Toby Ord argues predeployment testing misses the real risk: internal frontier models
ben_j_todd · x · 2026-11-12
Ben Todd relays Toby Ord's argument that the biggest loss-of-control and power-concentration risks come from internally deployed frontier models, so mandating predeployment testing does little — and may even worsen matters by widening the gap between internal and external models. Todd adds that predeployment testing mainly helps reduce misuse risks, which he sees as a secondary priority (though still potentially severe).
Related event: Toby Ord: biggest AI risks lie in internally deployed models(3 posts)→
More from AGI Musings
- Pedro Domingos coins 'Dario's law': compute needed for AI also grows exponentially — pmddomingos · 2026-11-12
- Matt Pocock: Clear business descriptions are half-built software in the AI era — mattpocockuk · 2026-11-12
- Researchers Push Efficiency-Gain Metrics Over Raw Benchmarks to Measure AI Progress — eliebakouch · 2026-11-12
- Bittensor Subnet NOVA Bets on Real Drug Assets to Prove Its AI Discovery Platform — markjeffrey · 2026-11-12
- Vercel CEO Rauch: great products still demand extraordinary focus and love — santiviquez · 2026-11-12
- Own your evals: dair-ai argues evals are the real moat of the intelligence stack — dair_ai · 2026-11-12