Draw on precedents from nuclear, climate, and chemical weapons governance to identify design principles for a credible international AI safety institution.

Comparative institutional analysis of IAEA, IPCC, CWC, CITES, and Basel Convention; structured expert interviews with 14 international law specialists; game-theoretic modeling of treaty participation incentives.

Three precedent institutions offer distinct design lessons for AI governance: (1) The IAEA demonstrates that verification-based regimes with on-site inspection rights can work when member states perceive mutual benefit from constraining all parties — but took 30 years to reach current effectiveness.

(2) The IPCC shows that intergovernmental scientific assessment panels can produce authoritative consensus without enforcement power, shifting the Overton window on acceptable national policies — however AI moves 100x faster than climate science, making IPCC-style timelines unsuitable.

(3) The Chemical Weapons Convention achieved 98% state party adherence through a combination of universality norms, industry verification, and clear red lines — most applicable to AI catastrophic misuse scenarios.

Key design requirements for an effective AI Safety Institution: (a) independent technical secretariat with hire/fire authority over domain experts, (b) incident reporting obligations analogous to ICAO aviation safety reporting, (c) capability threshold triggers for mandatory consultation (defined in compute/benchmark terms), (d) binding member state obligations for domestic AI safety frameworks, (e) sanctions authority for non-compliance.

Realistic timeline: a "Minimum Viable AI Safety Treaty" covering catastrophic misuse (CBRN assistance, critical infrastructure attacks) could achieve 30+ nation ratification within 3-5 years if negotiation begins by 2026.