OpenAI cofounder Zaremba calls for Vela-style verification infrastructure for AI
woj_zaremba · x · 2026-11-18
Wojciech Zaremba (OpenAI cofounder) invokes "trust, but verify" and the 1959 Project Vela satellite system that enabled nuclear test monitoring and subsequent treaties, arguing the AI era needs analogous technology.
He proposes technological solutions that make international policy agreements verifiable: confirming where chips are located, whether they're used for training, and whether AI is employed for cyber offense. He acknowledges the nuclear analogy is imperfect given AI's many positive applications, but argues high-stakes international agreements require such verification infrastructure — and humanity is in very early days of building an ecosystem for it.
More from AGI Musings
- FT: AI-tailored job applications create a lose-lose arms race for employers and applicants — paulnovosad · 2026-11-18
- Nina Schick: AI begins with the supply chain of intelligence, not models — NinaDSchick · 2026-11-18
- What if OpenAI or Anthropic leaked one week of chat logs? A thought experiment on data risk — dbasch · 2026-11-18
- MIRI President's Partner Runs Bay Area Kink Parties, Report Says — thedealdirector · 2026-11-18
- Turing's Point: A Machine Expected to Be Infallible Cannot Also Be Intelligent — wavefnx · 2026-11-18
- Pangram flags 99% of award-winning Peruvian novel as AI; publisher keeps prize — nabeelqu · 2026-11-18