Originally published in The National Interest.
As Washington and Beijing consider cooperation on AI safety, the United States must pursue transparency and risk reduction without undermining its technological leadership.
China and the United States may soon engage in diplomatic efforts to address mounting concerns over the bio- and cybersecurity risks posed by artificial intelligence (AI). Treasury Secretary Scott Bessent hinted at such a possibility at the conclusion of the recent summit between Presidents Donald Trump and Xi Jinping. While this development was largely left out of reporting on the gathering, it’s the latest sign of a pronounced shift in the Trump administration’s posture toward AI.
Rising AI Model Risks Are Forcing a Policy Reassessment
In the wake of Anthropic’s staged release of its Mythos model and subsequent evaluations of the bio- and cybersecurity risk introduced by the latest models, the Trump administration has been exploring how best to update its policy posture.
Since Day One of the administration, President Trump has explicitly endorsed a light-touch regulatory strategy. His aim has continually been to ensure that the United States develops and diffuses the most capable models. Vice President JD Vance and others have endorsed and implemented this policy vision.
The AI Action Plan set forth dozens of recommendations intended to continue the US’s AI leadership by reducing unnecessary or poorly tailored regulations. A more recent National AI Framework from the President outlined the steps Congress should take. This concern over state and federal laws that may hinder AI labs’ ability to drive AI innovation also manifested in an executive order that, in part, directed the Department of Justice to challenge laws that unduly interfered with the nation’s AI ambitions.
The demonstrated capabilities of the latest models indicate that officials within the administration may now see the AI landscape in a new light. Over the last few weeks, there have been pervasive rumors about executive action to subject leading models to some form of evaluation. The timing of such testing, the legal consequences of any findings, and the private and public stakeholders involved in any such process remain unclear. Bessent’s remarks indicate that the Trump administration is examining how to collaborate with China to ensure that any such highly capable models do not end up in the wrong hands.
Three Principles Should Guide Any US-China AI Cooperation
If the Trump administration proceeds with this updated AI policy stance, it ought to keep a few key principles in mind.
First, transparency related to model capabilities is a substantive step at this stage. An agreement between the United States and China to evaluate their models against common frameworks would provide valuable information to each nation. If and when such tests indicate any concerning signs, the nations could talk further on the basis of that new information. Additionally, more sensitive tasks, such as exchanging more detailed information about how the models were developed, are unnecessary until the aforementioned tests indicate a much higher level of risk.
Second, the President’s powers only go so far. Any domestic or international testing regimes will require some form of congressional approval. With respect to domestic requirements, Congress alone is in a position to develop a mandatory testing regime. With respect to the latter, the Senate will need to be involved in a formal treaty with China. It’s pivotal that the administration involves Congress and, by extension, the American people as it explores legislation and agreements that may have significant and long-term consequences.
Third, the White House cannot lose sight of the importance of the United States remaining the home of the world’s best AI. Up to this point, it has helped foster a vibrant and growing AI economy system. Those efforts must continue. The United States will be in the best position to guard against AI-related risks and harness its manifold benefits if it leads in AI innovation. Proposals—from DC or Beijing—that threaten the nation’s capacity to build the best AI should be greeted with immense scrutiny.
US AI Leadership Must Remain Non-Negotiable
US leadership in AI is the precondition for everything else on this agenda. Without it, American negotiators lose their leverage, their credibility in setting global standards, and their practical ability to understand the risks they’re trying to govern.
The principles laid out above exist to protect that position while still making meaningful coordination possible: Transparency first, Congress in the room, and no agreement that puts American labs at a disadvantage before the evidence demands it.