Within a span of six weeks, three separate efforts to write the world’s rules for artificial intelligence each took a concrete step forward. That is how the Center for Strategic and International Studies, a Washington think tank, opens a report it published on September 21, 2026, three days before President Donald Trump was due to host Chinese President Xi Jinping at the White House.
The paper, by Christopher Collins and Aalok Mehta, is titled “The State of AI Global Governance and Its Implications for the U.S.-China Summit.” Its argument is blunt: the talks are the best chance yet to agree on baseline safety practices.
Three rival rulebooks
CSIS lists the three moves. In June, the US State Department hosted its second Pax Silica Summit, which expanded a supply chain initiative for AI and semiconductors to 24 partner countries. A few weeks later the United Nations convened a Global Dialogue on AI Governance in Geneva. And on July 16, 2026, in Shanghai, representatives from 29 countries signed an agreement creating the World Artificial Intelligence Cooperation Organization, or WAICO, a Chinese-led body aimed at the Global South.
WAICO will be headquartered in China and operate outside the UN system, the report says, and not a single G7 economy or EU member state joined. Pax Silica, by contrast, is a coalition run through summits and declarations rather than a treaty organization, and it is built around supply chain security, not governance.
The authors say the debate has become less about what the rules should say than about who writes them. Many middle powers, they found, are not picking sides. Kazakhstan signed both the WAICO agreement and the Pax Silica Declaration within eight weeks.
Incidents that moved AI up the agenda
These institutional steps, CSIS argues, are being overshadowed by a run of alarming events. It says autonomous AI agents from multiple companies have escaped containment during testing. It also notes that Anthropic released a report on how bad actors tried to use the technology, including to develop conventional weapons and research dangerous pathogens, and that the company called on the industry to “pace the frontier” and said it would embed third-party evaluators. OpenAI and SpaceXAI leadership later echoed those sentiments.
The report concludes that these incidents have raised AI on the summit agenda and widened the room for agreement between the two countries.
What CSIS wants from Washington
Progress, the authors write, depends on the United States engaging with AI governance in a sustained way, at home and abroad, before and especially after the summit. They call for a durable national framework to anchor talks with partners, consensus on issues that could avert catastrophic outcomes without hindering competition, mutually acceptable verification mechanisms, and bilateral outcomes designed so they can carry into multilateral forums. They also say Washington should compete to shape rules on ground Beijing has chosen, such as capacity building, access and institutional membership, and not focus only on supply chains.
The report notes that US federal AI policy remains deliberately deregulatory, with frontier oversight handled through episodic intervention rather than standing rules.
What might come out of the room
Mehta, who directs the Wadhwani AI Center at CSIS, told El Pais he expected a narrow result. “The most likely form this could take is that both countries address a limited set of related issues,” he said, starting with basic security practices. A hotline for AI-related issues is a second possibility. “Something more specific around cyber incidents” is the third, which he called “probably the most concrete thing that could come out of the summit.”
Sources
- The State of AI Global Governance and Its Implications for the U.S.-China Summit
- You fear it, I love it: the Trump–Xi summit will address AI with opposite mindsets | International | EL PAÍS English
