OpenAI researcher: domestic compute-pact compliance is doable, international verification is the hard part
ohlennart · x · 2026-10-22
ohlennart weighs in on the feasibility of AI governance agreements: near-term domestic pacing looks doable, with good plans despite details to iron out.
The real challenge is the international component with meaningful verification — he calls it "a massive effort," noting that lots of work could make it less painful but it will remain a massive pain.
More from Safety
- Why the AI safety and security camps think so differently: mystics vs engineers — sebkrier · 2026-10-22
- Leaked PHP vulnerability forces early patch, massive attacks expected within 15 minutes — i_dg23 · 2026-10-22
- Pedro Domingos slams European AI research for treating personalization as a problem to combat — pmddomingos · 2026-10-22
- Railway breach tied to stolen Anthropic/OpenAI keys — API tokens are 'basically cash' on black markets — mitsuhiko · 2026-10-22
- New QUIC 0-day lets one connection OOM HTTP/3 servers; 7 of 10 libraries vulnerable — cephaloform · 2026-10-22
- GLM-5.3 Shows Unprompted "Update-Aware Reasoning," Apollo Finds 1-2 in 1,000 Rollouts — Don't Worry About the Vase (Zvi) · 2026-10-22