The rapid advance of “frontier” AI models after 2022 prompted a global shift toward multilateral governance focused on managing potential large-scale risks. International efforts, including the 2023 Bletchley Declaration and the United Nations General Assembly’s 2024 AI resolution, began to establish shared norms for AI safety and scientific cooperation. National policies initially mirrored this focus, often linking oversight to computational power. For instance, a 2023 US Executive Order established a template for compute-based governance, seeking to “Require that developers of the most powerful AI systems share their safety test results and other critical information with the U.S. government” (“FACT SHEET: President Biden Issues Executive Order on Safe, Secure, and Trustworthy Artificial Intelligence.” The American Presidency Project, 30 Oct. 2023). However, this federal approach changed after President Trump revoked Biden’s AI executive order in January 2025 and directed his administration to remove policies considered barriers to American AI innovation. Since then, US AI governance has become more uneven, with state legislatures taking a larger role in AI regulation while federal policy has prioritized AI leadership, competitiveness, and reduced regulatory barriers.

The authors of If Anyone Builds It, Everyone Dies view these measures as insufficient precursors to the robust international controls they deem necessary. Yudkowsky and Soares argue that preventing a catastrophic AI capabilities race would require internationally coordinated monitoring and restriction of the computing infrastructure used to develop advanced AI systems. Their position reflects a broader debate within AI governance about whether fragmented national regulations and voluntary corporate safety commitments can effectively constrain competition among governments and technology companies pursuing increasingly powerful models. From this perspective, while existing oversight mechanisms, reporting structures, and export controls provide useful groundwork, they fail to resolve the core incentive problem and remain inadequate for preventing continued escalation in frontier AI development.