As the technological rivalry between the United States and China intensifies, a new policy analysis suggests that artificial intelligence may be a rare domain where bilateral cooperation can succeed, precisely because the traditional tools of economic statecraft, such as tariffs and export controls, are proving ineffective.
The analysis, authored by Charlie Andrade and recognized as the runner-up for the Jimmy Carter Prize in the China Focus Essay Contest, argues that the window for meaningful AI cooperation is currently open but narrowing rapidly. The core premise is that unlike broad manufacturing sectors targeted during the trade war, the frontier AI industry is highly concentrated, making targeted regulatory agreements feasible.
The essay cites the November 2024 Lima agreement between US President Joe Biden and Chinese President Xi Jinping, which stipulated that decisions regarding nuclear weapons should be made by humans, not AI, as a foundational precedent. Andrade characterizes this as “more than the two governments had agreed on any other shared problem in fifteen years.”
The Failure of Broad Decoupling
The analysis contrasts the potential for AI cooperation with the largely unsuccessful attempts to decouple the broader US-China economic relationship. From 2017 to 2024, while China’s share of US direct imports fell by 8 percentage points, total US merchandise imports continued to rise as firms simply shifted final assembly to countries like Vietnam, Mexico, and Taiwan, often relying on Chinese components.
Similarly, a study of Japanese multinationals in China between 2012 and 2022 found that while 81% adopted at least one “de-risking” strategy, only 18-20% actually decoupled. A significant 43% expanded operations elsewhere while leaving their Chinese footprint entirely intact.
Even in the highly scrutinized semiconductor sector, broad restrictions are facing pushback. The Semiconductor Industry Association, representing 99% of the US chip industry by revenue, has publicly warned that expansive export controls risk “harming the US semiconductor ecosystem without advancing national security.” Furthermore, China’s control over approximately 70% of global rare-earth mineral supply provides significant leverage against aggressive decoupling efforts.
A Narrow Window for AI Auditing
Because broad economic decoupling is failing, the essay argues that policymakers must pivot to highly targeted cooperation in critical areas like AI safety. The regulatory target for frontier AI is remarkably small—a “state-adjacent set of actors” comprising perhaps two dozen organizations globally, including OpenAI, Anthropic, and Google in the US, and Baidu, Alibaba, ByteDance, DeepSeek, Moonshot, and Zhipu in China.
Both nations are already building the domestic infrastructure for such regulation. The US established its AI Safety Institute (now the Center for AI Standards and Innovation) in late 2023, while the China AI Safety and Development Association was announced in 2024. Notably, China issued as many national AI safety standards in the first half of 2025 as it had in the previous three years combined.
The proposed framework for cooperation centers on third-party auditing of AI training runs that exceed a specific compute threshold. Because the industry is currently dominated by a handful of players, establishing a bilateral auditing regime is logistically possible now, before the technology proliferates further.
As Washington probes US firms using Chinese AI models, the analysis suggests that establishing shared safety standards, rather than relying solely on export controls, may be the only viable path to managing the risks of frontier artificial intelligence.
Obstacles and the Narrowing Window
The essay is candid about the significant obstacles to any bilateral AI cooperation framework. The most fundamental is the deep mutual distrust that has accumulated over years of technology rivalry. US policymakers are acutely aware that any information shared in the context of a safety-auditing regime could be used to accelerate Chinese AI development. Chinese officials, for their part, are wary of any arrangement that could be used to impose external constraints on domestic technology development or provide the US with intelligence about Chinese AI capabilities.
The political environment in both countries also complicates the picture. In the United States, any initiative that could be characterized as “cooperating with China on AI” faces significant bipartisan skepticism. In China, the framing of AI development as a matter of national sovereignty and technological self-reliance makes formal concessions to foreign oversight politically difficult.
Despite these obstacles, the essay argues that the window for cooperation is currently more open than it appears. The shared interest in preventing catastrophic AI failures, whether from misaligned systems, weaponized AI, or accidental escalation, provides a genuine foundation for dialogue that transcends the competitive dynamics of the broader technology rivalry. The Lima agreement on nuclear AI, however limited, demonstrates that such agreements are achievable even in the current political climate.
The analysis ultimately frames the choice as a binary one: either the two countries establish some form of shared governance framework for frontier AI before the technology proliferates beyond the current small number of actors, or they face a future in which the risks of uncoordinated AI development are managed, if at all, through the same blunt instruments of tariffs and export controls that have proven so ineffective in the broader technology competition.