ICAS Trade ‘n Tech Dispatch (online ISSN 2837-3863, print ISSN 2837-3855) is published about every two weeks throughout the year at 1919 M St NW, Suite 310, Washington, DC 20036.
The online version of ICAS Trade ‘n Tech Dispatch can be found at chinaus-icas.org/icas-trade-technology-program/tnt-dispatch/.
Recent disputes over AI model access and model distillation suggest that U.S.-China competition is entering a new phase. Beyond technological rivalry, Washington and Beijing are increasingly advancing different approaches to governing AI development, international cooperation, and access to emerging technologies—differences that could shape the future of global governance far beyond artificial intelligence.
In One Sentence
Mark the Essentials
Keeping an Eye On…
To the surprise of absolutely no one, U.S. Trade Representative Jamieson Greer announced on July 1 that the Trump administration would not renew the USMCA in its current form. The decision ensures that the agreement will now enter a cycle of annual reviews, with a hard expiry date of 2036 if the three countries do not reach a final resolution. Keeping alive the threat of the agreement’s expiration is intended to extract further concessions from Mexico City and Ottawa so as to lower the bottom-line metric that has all along driven the administration’s trade policy — namely, reducing the gaping bilateral goods trade deficits. The deficit with Mexico reached $197 billion in 2025; that with Canada, $48 billion — a 36 percent increase in real terms from 2019 to 2025, overall.
The auto sector deficit is, if anything, proportionately larger, especially with Mexico, having grown from $91 billion in 2019 to $131 billion in 2025. Heavy truck imports and auto parts imports have in particular driven up the deficit numbers with Mexico — the former quadrupling from $3.8 billion in 2019 to $17.4 billion in 2025; the latter increasing from $59.3 billion in 2019 to almost $80 billion in 2025. Worse, U.S. content in Canadian- and Mexican-produced autos has also declined while Chinese auto part content has steadily risen, albeit from a low base. Cue the call for further modification of auto rules of origin (ROOs) to elbow out Chinese content, especially electronic components such as semiconductors, circuit boards, and displays (fully assembled Chinese automobiles have already more or less been elbowed out of the U.S. marketplace via sky-high tariffs and stringent connectivity and vehicle software rules). The dependence on these inputs, especially in newer technologies such as electric and autonomous vehicles, would presumably pose a strategic and economic resilience risk.
As such, one should expect the administration to press its North American partners to modify the USMCA’s core auto-parts list to include electric drive unit (EDU) production and to further tweak the labor value content (LVC) rule to ensure that such EDU production takes place on U.S. soil. It bears remembering that the USMCA’s automotive ROOs were already an upgrade — or, rather, a protectionist downgrade — from NAFTA’s ROOs: the regional value content (RVC) was upped from 62.5 percent to 75 percent; a labor value content (LVC) rule was introduced requiring that 40–45 percent of a vehicle’s production by value be made by workers earning at least $16 per hour; and a steel and aluminum origination requirement was introduced as well. Expect the RVC and LVC requirements to be raised even further, and a U.S. domestic production requirement within the North American bloc to be added during the ongoing joint review negotiations.
Beyond the auto sector, what are the U.S.’s other key demands? There are many. Both Mexico and Canada must strengthen ROOs for industrial goods to minimize the use of third-country content in U.S. supply chains. Second, both countries must enhance their economic security alignment on tariffs, export controls, and foreign investment screening. Third, they must align their regulatory frameworks to prevent the offshoring of U.S. production to Mexico and Canada. Fourth, they must improve implementation of their forced labor import bans. And fifth, they must assist in the development of a North American critical minerals marketplace to minimize the potential for Chinese weaponization of such dependencies.
Over and above these joint demands, Mexico must also improve its labor and environmental law enforcement, address U.S. concerns regarding certain energy policies and practices, address U.S. concerns regarding the methodology it applies to calculate spectrum user fees, treat U.S. electronic payment service suppliers equally so that they can process domestic transactions using their own proprietary networks, and accommodate the impact of imports of Mexican seasonal produce on U.S. growers. For its part, Canada must provide greater market access for U.S. dairy products, address the impact of its Online Streaming and Online News Acts on U.S. digital service providers, and address provincial bans on the distribution of U.S. alcoholic beverages as well as various discriminatory government procurement measures in certain provinces.
Don’t be distracted by this laundry list of demands. At the end of the day, it will be the bilateral trade deficits with its two North American partners that ultimately inform the U.S.’s negotiating strategy, at least over the next two review cycles. The Canadians have an easier hand to play in this regard, given that if one excludes Ottawa’s energy and metals exports — already hard-wired into its trading relationship with Washington — Canada in fact runs a goods trade deficit with the U.S. Perhaps this also partly explains Prime Minister Carney’s tendency to talk back to the American side, in contrast to President Sheinbaum’s studied reticence. Unfortunately for Sheinbaum, she will have to hold her tongue a lot longer, given that Mexico’s USMCA-dependent automotive sector represents close to 5 percent of Mexico’s GDP and employs hundreds of thousands of workers — precisely the sort of leverage the American president salivates over.
Expanded Reading
U.S. Declines to Renew U.S.M.C.A., Starting 10-Year Clock to Expiration, The New York Times, July 1, 2026
U.S. auto industry faces increased uncertainty without extension of USMCA trade deal, CNBC, July 1, 2026
Mexico’s Export-Led Economy at Risk With Annual USMCA Reviews, Bloomberg, July 2, 2026
Canadian dollar heads for steep monthly decline ahead of USMCA deadline, Reuters, June 30, 2026
Trump Halts USMCA Trade Agreement: How Grocery Bills and Car Prices Could Change, Newsweek, July 2, 2026
In One Sentence
Mark the Essentials
Keeping an Eye On…
The Trump administration is caught on the horns of a dilemma when it comes to regulating the increasingly powerful AI models that now seem to drop every other month, or even every week.
On one hand, the administration is ideologically committed to a laissez-faire approach to model regulation so as not to impede innovation within the frontier AI space. On the other hand, the cybersecurity and biohazard risks inherent in frontier AI capabilities are too grave to ignore. The messy compromise has been an executive order (EO) that creates a classified benchmarking framework to assess the advanced cyber capabilities of frontier AI models but leaves AI labs’ adherence to the framework’s processes voluntary. As the EO explicitly notes, the classified benchmarking framework must not be construed as creating a mandatory governmental licensing, preclearance, or permitting requirement.
On one hand, the administration is reflexively committed to an America First approach to AI model regulation that prioritizes domestic innovation within the U.S. ecosystem, as well as global enforcement subject to U.S. jurisdiction. On the other hand, “America First” amounts to “America Alone” in this instance, given that U.S. allies and partners have no influence over or voice within the deployment of controls over U.S. frontier AI models, and in fact have had their access to these models cut off without preparatory consultation or notice. The messy workaround toward a more consistent framework of reliable access is likely, over time, to feature a tiered, positive-list approach to model release — one that would see the U.S. government and select U.S. firms granted access first, followed by other U.S. firms and top allies and partners. In many ways, this tiering could come to resemble the categorization of countries under the Biden-era AI “Diffusion Rule,” which Trump ostentatiously junked early in his term.
On one hand, the administration is wholly committed to exporting the entire U.S. AI stack to foster usage, dependency, and even addiction to the stack globally. On the other hand, such unchecked access will inevitably empower competitors and adversaries as they “distill” U.S. model capabilities to train, upgrade, and perfect their own models. The messy compromise in this instance is likely to feature a mix of export controls — based on model weights as well as the geographic location of end use — and protocols and standards designed to inform and erect guardrails during the process of advanced model releases. To be clear, this is not a dilemma that confronts the U.S. alone; Chinese regulators are understood to be considering limiting overseas access to their own advanced AI models, including open-weight models, both from a competitiveness standpoint and from a security, technology management, and responsible-use perspective.
One thing is becoming clear, though: on both sides of the Pacific, regulation has begun the long journey of catching up with the technology. A less ad hoc and more prescriptive approach to regulation is very much on the cards. An important question arises in this regard: will this new prioritization of regulation also lead to a splintering of what is today still, for the most part, an undivided global AI ecosystem featuring an American stack and a Chinese one? Or will the emergence of two ecosystems with very limited touchpoints become an assured inevitability? It would not be an exaggeration to say that much about the future of AI — and of the Fourth Industrial Revolution more broadly — will hinge on the answer to this question.
Expanded Reading
Legislative Developments
Hearings and Statements
Expanded Reading
Artificial intelligence has become the latest focal point of U.S.-China strategic competition—not because of another breakthrough model, but because of a growing dispute over who should have access to advanced AI and how that access should be governed.
The latest controversy surrounding Anthropic illustrates this shift. Reports revealed that the company had experimented with identifying China-based users of its coding assistant Claude Code while simultaneously accusing several Chinese AI companies of using “model distillation” to narrow the technological gap with leading American systems. What might once have been viewed as a commercial dispute over intellectual property has increasingly been framed as a matter of national security.
Neither of the underlying issues is particularly new. Model distillation has long been a common engineering practice across the AI industry, and American companies have also relied on similar techniques. Companies have likewise always sought to protect proprietary models from unauthorized use. What has changed is not the technology itself, but the strategic context in which these practices are now viewed. As U.S.-China AI competition intensifies, commercial concerns are increasingly being redefined as strategic ones. If this sounds familiar, it should. Similar transitions have already transformed debates surrounding Huawei, TikTok, connected vehicles, and a growing list of emerging technologies. The technologies of concern change, but the policy logic remains remarkably consistent.
Recent developments suggest that AI is no longer being governed primarily as an innovation ecosystem, but increasingly as a strategic one.
In the United States, the Trump administration has steadily expanded AI policy beyond innovation and technological leadership toward national security and strategic competition. Executive actions have tightened restrictions on foreign access to frontier AI models, while closer coordination between government agencies and leading AI firms has elevated model security into a national policy priority. Treasury Secretary Scott Bessent recently summarized this evolving mindset bluntly, arguing that “the biggest risk to AI is China getting ahead of us.” Viewed from this perspective, controlling the diffusion of frontier AI capabilities becomes not simply a commercial issue, but an instrument of national security.
China, meanwhile, has pursued a noticeably different approach. Alongside proposals for a global AI cooperation organization and continued support for multilateral AI governance, Beijing has consistently framed AI as a driver of industrial upgrading, digital transformation, and broader economic development. While Chinese companies compete aggressively at the technological frontier, China’s AI ecosystem has generally placed greater emphasis on open-weight models, lower-cost deployment, and wider commercial adoption. Rather than allowing market demand alone to shape AI development, Beijing has integrated AI into broader national development objectives from the outset.
The difference reflects more than competing AI strategies. The emerging U.S. approach begins with security, emphasizing trusted networks, managed access, and selective participation. China’s approach begins with development, emphasizing technological diffusion, international cooperation, and broader participation. Both are designed to strengthen national competitiveness, but they increasingly represent different models for organizing international technological cooperation.
The implications extend well beyond artificial intelligence. Similar policy dynamics have already emerged across investment screening, export controls, industrial policy, supply chain security, and market access. Over the past decade, Western economic policymaking has gradually evolved from prioritizing innovation, to securitizing economic interdependence, and now increasingly toward governing access. AI has simply become the clearest manifestation of this broader transformation.
This evolution also helps explain why international economic cooperation has become increasingly difficult even where commercial interests remain aligned. Chinese overseas investment is often framed by Beijing as supporting industrial development, employment, and economic modernization, while policymakers in Washington and parts of Europe increasingly evaluate many of these same activities through the lens of technology transfer, ownership structures, strategic dependence, and long-term security implications. The disagreement is therefore no longer limited to individual technologies or investment projects—it increasingly reflects competing assumptions about the purpose of international economic engagement itself.
For many countries, however, neither vision fully reflects their own priorities. Most developing economies are less interested in joining technological blocs than in gaining access to affordable AI tools, digital infrastructure, investment, and innovation that support their own development strategies. The emerging competition over AI therefore raises a broader question that extends beyond technology itself: whether the governance of emerging technologies will evolve into competing international systems that redefine the rules of global cooperation. If so, the greatest challenge may not be technological fragmentation, but the gradual fragmentation of global governance.
This issue’s Spotlight was written by Yilun Zhang, Research Associate at ICAS.