Governing frontier AI under U.S.–China competition: a staged hybrid model for reducing escalation risk

Frontier artificial intelligence (AI) creates escalation risks under U.S.–China competition because rapid technological change, strategic secrecy, and private control of advanced systems complicate effective oversight. This paper evaluates which governance model can best reduce these risks through a qualitative comparative case-study approach supported by legal analysis. It compares binding multilateral law, expert soft law, bilateral agreements, the fragmented existing AI regime, and hybrid governance through the WTO, Tallinn Manual, SALT I and the ABM Treaty, current AI instruments, and outer-space governance. The comparison suggests that no single model adequately combines adaptability, participation, verification, and enforcement. Binding law is most useful for narrow, observable obligations; soft law supports technical clarification and experimentation; and bilateral coordination can address shared vulnerabilities between strategic rivals. The paper therefore proposes a four-stage hybrid framework: domestic oversight of frontier developers, a permanent U.S.–China commission, tested nonbinding safeguards, and a narrow binding treaty focused on nuclear decision-making and serious incident notification. Broader participation would follow as the framework becomes operational. By linking international commitments to domestic regulatory capacity and making legal obligations conditional on practical verification, this work identifies a feasible path for reducing miscalculation and technical-failure risks without assuming that governance can eliminate geopolitical competition.