Rahul Somvanshi
The question worth asking before the Trump-Xi meeting is not whether the US and China can sign a sweeping AI treaty. It's whether they can agree on a far narrower set of rules before the competition produces a crisis no one intended.
As the two governments prepare for September 2026 talks, AI governance has emerged as a prominent issue in discussions surrounding the planned Trump-Xi meeting, alongside trade, technology and chip controls. The pressure to establish some kind of shared framework is growing from multiple directions, including international bodies and research institutions that have spent years trying to define what agreement is even possible.
The immediate backdrop is a competition that neither side shows signs of slowing. The US has restricted advanced chip exports to China while pushing domestic AI development. China has accelerated its own investment. Both countries are developing and integrating AI for military applications, while researchers and policymakers are debating whether governance mechanisms are keeping pace.
What a narrow agreement could actually look like
The most concrete proposals in circulation do not attempt to govern the entire AI sector. They focus on military AI — specifically, where humans must stay in control and where AI-enabled systems should not be allowed to operate without human authorisation.
Researchers involved in a Brookings-Tsinghua Track II dialogue have proposed that both governments commit to keeping humans responsible for any decision involving attacks on nuclear command systems and critical infrastructure. They also propose establishing crisis-management mechanisms so that if autonomous systems behave in unexpected ways, the two sides have a direct channel to communicate before a situation escalates.
Those proposals are not yet government policy on either side. They are proposals from researchers for a narrower set of safeguards that could provide a starting point for bilateral discussions.
UN Secretary-General António Guterres has pressed for AI guardrails at the international level, calling for frameworks that are safe, transparent and accountable. The UN has established an Independent International Scientific Panel on AI to provide scientific evidence to support AI governance discussions. Neither mechanism is a substitute for direct agreement between Washington and Beijing, but both reflect the scale of pressure on governments to act.
Researchers argue that both countries have an interest in reducing the risk of accidental escalation involving AI-enabled military systems. That shared logic is precisely why military AI safeguards — rather than broader AI governance — may be the only place where real progress is possible in the short term.
What both sides have not agreed on
The harder questions concern everything beyond crisis communication and nuclear-adjacent systems. AI development timelines, data access, model training, civilian applications and the definition of "dual-use" AI remain contested at every level. Chip controls are a continuing source of tension rather than a foundation for cooperation.
The September meetings are unlikely to resolve all of those deeper disagreements, particularly where the two governments' technology and security interests diverge. What they might produce is an acknowledgment that military AI, at minimum, requires human control over the most consequential decisions. That would be a starting point, not a solution — but a starting point is what the current situation lacks.
Further reporting will clarify the governments' positions as official statements emerge.
