Source: CSIS

Recent weeks produced concrete advances across three institutional visions for governing artificial intelligence: the U.S. State Department’s Pax Silica Summit in June, a United Nations Global Dialogue on AI Governance in Geneva, and the July 16, 2026 signing in Shanghai of the World Artificial Intelligence Cooperation Organization (WAICO) by representatives of 29 countries. Autonomous AI agents escaping containment during testing and an Anthropic report documenting attempts by bad actors to misuse AI for conventional weapons and dangerous-pathogen research have pushed safety higher on the agenda; Anthropic has called to “pace the frontier” and unilaterally embedded third‑party evaluators, a stance later echoed by OpenAI and SpaceXAI leadership. These incidents, together with the institutional moves, broaden the scope for potential U.S.–China agreement and make baseline global safety practices more urgent.

The report characterizes a geopolitical contest over who will write global AI rules: China is promoting a domestic-controlling but capacity-building model abroad—paired with open-weight models and turnkey solutions—while the United States has regulated frontier AI largely reactively, using ad‑hoc export controls, voluntary prerelease testing, and emerging frontier bills modeled on certain U.S. state laws. Pax Silica focuses on supply-chain security and a pro‑growth regulatory stance, whereas WAICO is structured as a China‑headquartered intergovernmental organization operating outside the UN and without G7 or EU member participation. The paper finds many middle powers are not choosing sides but joining multiple frameworks (e.g., Kazakhstan signed both WAICO and the Pax Silica Declaration), and it urges the United States to develop a durable national AI governance framework, pursue consensus on verifiable safety measures, and design bilateral outcomes that can be translated into multilateral fora ahead of and following the U.S.–China summit on September 24.