Frontier supercomputer, ORNL archive. Carlos Jones/ORNL, U.S. Dept. of Energy. Public domain; cropped.
President Donald Trump and Chinese President Xi Jinping head into their scheduled Shenzhen meeting with two different artificial-intelligence questions: how to warn each other when advanced systems cause serious trouble, and who gets access to the computing power that makes those systems possible.
The first is an area where Washington has proposed cooperation. The second remains a national-security and commercial dispute. Progress on one would not automatically settle the other.
The White House travel schedule distributed to the press pool lists the bilateral meeting for Thursday, November 19, at 12:45 p.m. in Shenzhen, or 11:45 p.m. Wednesday in New York. A trilateral meeting with Xi and Russian President Vladimir Putin is scheduled afterward. Those are planned sessions, not completed talks or agreements.
The AI promise already on the table
Trump and Xi did not arrive at this week’s summit without an existing framework. In its September 25 account of Xi’s Washington visit, the White House said the countries had established a Super Intelligence Dialogue to discuss the technology’s risks and benefits, with another exchange due by November. They also agreed to establish a bilateral communications channel for incidents involving the technology.
The administration uses “super intelligence,” or SI, in those statements. For readers following the broader AI debate, the practical issue is what governments do when increasingly capable computer systems create risks that cross national borders.
In remarks published November 11, Treasury Secretary Scott Bessent described three priorities: sharing practices for evaluating advanced models before release, keeping human judgment in command, and creating a direct channel through which either country could alert the other to serious incidents involving its models.
That sets a more concrete test for the talks than a general declaration that AI should be safe. An alert system needs people responsible for receiving a warning, a shared understanding of what should trigger one, and procedures for responding. A promise to communicate is a starting point; it does not, by itself, establish those operational arrangements.
Testing before release, not only talking after an incident
An incident channel concerns what happens when a serious problem emerges. Testing before a model is released addresses an earlier stage: trying to identify dangerous capabilities or failures before the system reaches users.
Bessent’s stated goal is an exchange of best practices. That is different from an agreement to use identical tests, allow mutual inspections, share companies’ underlying technology or stop developing more capable systems. Readers assessing any announcement should look for the specific measures agreed, which organizations will carry them out and whether a timetable follows.
Keeping human judgment in command raises a related implementation question: where decisions remain with people, and how that control is maintained as systems take on more tasks. A shared statement of principle would not alone explain how either government or private companies must put it into practice.
Cooperation on safety does not mean open access to chips
While preparing for dialogue, Washington has also moved to restrict access to advanced American computing. On Monday, November 16, the Commerce Department’s Bureau of Industry and Security announced an interim final rule addressing remote access to advanced chips and servers through data centers in third countries.
The agency said the rule was designed to prevent existing technology restrictions from being bypassed through remote computing arrangements. The distinction matters because a company can use computing capacity over a network without physically importing the chips that provide it.
That creates separate stakes for chipmakers, cloud providers and businesses building AI services. Safety discussions concern how powerful systems are tested and how governments handle serious incidents. Computing-access restrictions concern which customers can use the hardware and services needed to develop and run those systems.
Agreement on an incident channel would therefore not, on its own, authorize new chip shipments or remove restrictions on remote access. Those changes would require their own decisions. Nor should businesses treat a diplomatic statement as a substitute for the applicable export-control rules.
What would count as a useful result?
The clearest evidence of progress would be practical detail: designated contacts for serious incidents, an agreed process for sharing warnings, a defined program of technical exchanges or a schedule for follow-up work on model evaluation.
Those would be steps toward implementing the earlier promise, not proof that AI risks had been eliminated or that the U.S.-China technology rivalry had ended. For companies and the people using their systems, the important distinction is between leaders endorsing cooperation and institutions being given specific work to do.
JBizNews Desk | Washington
© JBizNews.com. All rights reserved. Unauthorized reproduction or redistribution is strictly prohibited.



