We Need AI Treaties. This is How We Get Them

Effective AI governance requires developing verification technologies that enable international cooperation, similar to the frameworks used for nuclear non-proliferation. Tim Fist and Janet Egan emphasize that because AI development relies on a highly concentrated compute supply chain—dominated by a few chip designers and cloud providers—it is possible to implement monitoring mechanisms that ensure compliance without requiring total trust between adversaries. Key strategies include utilizing Trusted Execution Environments for cryptographic attestation, implementing location verification for hardware, and retrofitting data centers to monitor usage patterns. By shifting from performance-based export controls to conditional frameworks that incentivize governable hardware, nations can establish verifiable red lines. These technical solutions, supported by human inspections and international oversight, provide a pathway to manage existential risks like recursive self-improvement and cyber-offensive capabilities, even in low-trust geopolitical environments.