Frontier AI & National Security Experts - Governing AI Without Trust

Can we build a global AI safety regime when the U.S. and China do not trust each other? In my new Fortune piece, I argued that the answer is yes. In fact, that should be the starting point for the model. Most discussions of international AI governance start with the question of how to get countries to agree on shared principles. I think that may be the wrong question. The U.S. and China are unlikely to agree anytime soon on privacy, surveillance, censorship, military use, or the values advanced AI systems should serve. They do not need to agree on all of that. They need to agree on a small number of scenarios neither side wants to see: systems capable of autonomous replication, accelerating the development of their own successors, evading oversight, or enabling catastrophic biological or cyber harm. In my new #Fortune piece, I propose a different approach to AI governance: Don’t build it on trust. Build it to work when trust is absent. The model rests on three elements: agreed early warning indicators, independent technical verification, and real consequences when commitments are violated. It draws on lessons from another system I spent years working in: the global fight against money laundering, terrorist financing, and proliferation financing. Countries in that system did not always trust one another. It worked because trust was not a prerequisite. What mattered were agreed rules, professional assessment, and consequences. The Fortune piece is a first attempt to translate that logic to frontier AI. I’d especially like to hear from people working on frontier AI, AI safety, national security, compute governance, and international coordination. There are still hard design questions to solve, and I’m now working on turning this into a much more concrete model. Governing AI Without Trust #AIGovernance #AISafety #FrontierAI #ArtificialIntelligence #Geopolitics #recursive