A Summit with High Stakes for AI Policy
With President Trump preparing to meet Chinese President Xi Jinping in Beijing, two of Washington’s most prominent think tanks have published analyses of what the United States should, and should not, do on artificial intelligence at the summit. The Council on Foreign Relations (CFR) and the American Enterprise Institute (AEI) reach broadly similar conclusions about the danger of concessions, but they frame the problem differently and prescribe different strategies.
Both articles were published in the days immediately before the summit, and both are directed at the same audience: American negotiators who will be sitting across the table from Chinese counterparts expected to demand the relaxation of US export controls on advanced chips as a precondition for any AI dialogue.
The CFR Position: Maximum Pressure, Targeted Dialogue
The CFR analysis, written by Chris McGuire, a senior fellow for China and emerging technologies who previously led US-China AI policy at the National Security Council under Biden and served as the State Department’s lead on US-Russia arms control during the first Trump administration, was published on May 9, 2026 under the title “How Trump Should Approach AI Talks with China: Targeted Dialogue, Maximum Pressure.”
McGuire’s central argument is that the United States currently holds a significant advantage in AI that it should not bargain away. China is approximately eight months behind the United States in AI development, a gap that China believes it can overcome. More strikingly, China will produce only about 2 percent of the AI computing power of US firms this year. That asymmetry is the product of years of export controls, and it is the primary source of American leverage.
The CFR article identifies three options available to the Trump administration. The first is to loosen export controls and hope that China negotiates in good faith. This is an approach McGuire dismisses as naive, citing the 2024 Biden-era US-China AI dialogue in which the US sent technical experts while China sent diplomats who complained about export controls. The second option is maximum pressure: tightening export controls further to expand the US lead from eight months to a potential 18-24 months. The third is to wait for a catastrophe before acting. McGuire recommends the second option.
The CFR article also catalogs the ways China is working around existing controls: purchasing US chips through intermediaries, accessing US AI capabilities through remote cloud services, smuggling chips via third countries, and using US chipmaking technology to advance domestic production. These loopholes, McGuire argues, make it all the more important to tighten controls rather than relax them.
On the question of dialogue, McGuire does not oppose it entirely. But he argues that any dialogue must be narrowly scoped to safety and risk reduction, specifically preventing AI-enabled military escalation, and must be paired with maximum pressure on export controls. Trump’s goal in Beijing, he writes, should be to “create the conditions for such an agreement down the road,” not to reach an agreement at the summit itself.
The AEI Position: Coordination Without Illusion
The AEI analysis, written by Ryan Fedasiuk, a fellow whose recent work has focused on export controls and China’s AI chip ecosystem, was published on May 7, 2026 under the title “AI Coordination Without Illusion.” It takes a more nuanced position than the CFR piece, arguing that some form of US-China AI coordination is not only possible but necessary, while being clear about what it should not involve.
Fedasiuk’s starting point is that the Trump-Xi summit arrives at an unusual moment. Four weeks after the release of Claude Mythos, a frontier AI model that Fedasiuk describes as having crossed “a frightening, existential threshold” in its capacity to discover and exploit software vulnerabilities, both the US and Chinese governments are, for the first time, taking AI safety seriously as a national security concern rather than a regulatory nuisance.
The AEI article frames the debate in Washington as splitting into two camps. The first wants to recreate the US-Soviet arms control architecture for AI, with hotlines, working groups, and verification regimes. The second views any conversation with Beijing about AI as appeasement. Fedasiuk rejects both positions. The first underestimates how different AI is from nuclear weapons and how different China in 2026 is from the Soviet Union in 1972. The second is wrong to conclude that no useful conversation with China is possible.
What Fedasiuk proposes instead is “coordination,” parallel domestic commitments by two governments that distrust each other, applied to a narrow set of shared risks neither can manage alone. He identifies three specific commitments worth pursuing at the summit: first, parallel commitments by both governments to facilitate pre-release technical evaluation of frontier models by a designated national authority; second, coordination on DNA synthesis screening to prevent AI-assisted creation of pandemic pathogens; and third, an ongoing channel for discussing AI-related cyber incidents, modeled on the Military Maritime Consultative Dialogue.
On export controls, Fedasiuk’s position aligns with McGuire’s. He writes that export controls on advanced chips are “a rapidly appreciating asset” and that “trading them away in 2026 for the promise of dialogue would risk echoing the very same mistake with China” that produced a decade of broken promises. Beijing is expected to demand chip control relief as a precondition for dialogue; Fedasiuk argues that American negotiators must refuse.
Where the Two Analyses Diverge
The most significant difference between the CFR and AEI pieces is in their framing of what dialogue can accomplish. McGuire is skeptical that Beijing will negotiate in good faith on AI safety at all, and his three-option framework essentially treats dialogue as a means of buying time while tightening the technological vice. Fedasiuk is more optimistic about the value of coordination, arguing that even a soft bilateral commitment to discuss AI security will trigger internal coordination inside both governments and create competitive pressure on China to appear more responsible.
Both articles, however, converge on the same bottom line for the summit: do not trade export control concessions for AI dialogue. The chip controls are leverage, not a bargaining chip. And the summit’s value lies not in any grand agreement but in the pressure it creates on both governments to define, on paper, what they are actually prepared to ask of each other.
As EastFrontier has previously reported, the US has been tightening its AI chip export control regime throughout 2026, with the Senate introducing the MATCH Act to block AI chipmaking equipment sales to China and the House Foreign Affairs Committee marking up 23 bills targeting China’s AI and chip supply chains.