What happened
RAND researchers built a two-stage game-theoretic model to identify when rival states can sustain mutual restraint in AGI development. The central finding is that 'without domestic AI oversight, rivals can't credibly commit to restraint — even when both sides want to': any restraint arrangement depends on verification, and verification depends on states having built domestic institutions (compute registries, licensing, facility audits, incident reporting) that make their own AI ecosystems legible. The report further finds that verification loses force as AGI arrival appears imminent, that fear of shared catastrophe deters some but not purely positional defections, and that confidential-computing and trusted-hardware verification methods can lower the security cost of demonstrating compliance.
Why it matters
Reframes domestic AI governance from regulatory compliance into security architecture that could enable US-China restraint arrangements — directly relevant to national-security and geopolitics positioning ahead of AGI-relevant negotiations.
Action needed
If you advise on AI national-security policy, treat compute registries and auditing capacity as verification infrastructure rather than compliance overhead, and track confidential-computing verification methods.