Talking to Beijing about AI safety is not the scandal. During the Cold War, Washington and Moscow negotiated arms control while loathing each other, and in 2024 Joe Biden and Xi Jinping agreed that humans, not algorithms, should control nuclear weapons. If the two countries building the most powerful AI systems on earth can agree on anything, that is worth having. The scandal is who is doing the talking, and what he has done at home.
Trump says he and Xi agreed that "at least minimal" safeguards would be a good thing, and that they never defined them. That is a sentiment, not a policy. It is also the same position he has taken toward American voters, whom he has asked to trust the industry to police itself.
Back in office, Trump dismantled the Biden-era approach to AI oversight, which was modest and mostly relied on voluntary reporting. His administration's AI plan treats regulation as an obstacle to winning a race. He backed a push to bar states from writing their own AI rules, and the Senate stripped that provision from his tax bill by a vote of 99 to 1. When 99 senators agree on something, it is usually because their constituents are already furious. Trump has since pursued the same goal by executive order. Washington has passed no serious AI law, so the states are the only governments acting for ordinary people, and he is trying to take that power away from them.
So the pattern is plain. Abroad, he will discuss safeguards with a fellow strongman over a friendly meeting. At home, he treats safeguards as sabotage. The only guardrails he seems to accept are ones negotiated privately between two heads of state, with no legislature, no courts, no workers, no parents, and no public in the room.
Then there is Xi's language. He wants the technology to "remain under human control." In Beijing, that has a specific meaning: the Communist Party's control, including control over speech and surveillance. A bargain that defines "safety" as keeping power in the hands of whoever already holds it is not a safeguard for the rest of us. And an American president who calls Xi and Putin "friends" in the same breath is not well placed to argue about whose control we should fear.
The contradictions go further. The administration has loosened restrictions on selling advanced Nvidia chips to China, at the same time it claims an AI arms race justifies deregulating American companies. The race is an emergency when someone asks Silicon Valley to submit to oversight. It is a business opportunity when Silicon Valley wants to sell abroad. The consistent principle is that industry gets what industry wants.
Democrats should not feel smug, because this is an opening they have been slow to use. Too much of the party's AI politics is reactive, a list of complaints about deepfakes and job losses with no clear agenda. Voters already worry about what these systems will do to their jobs, their children and their electric bills. A Democratic position should be specific: independent testing before release, liability when systems cause harm, whistleblower protections for lab employees, and public say over the data centers that are reshaping local power grids. None of that needs Xi's blessing. It needs a government willing to say that powerful technology answers to the public.
If Trump wants a real conversation with China about AI, fine. Start with something that can be verified, such as a ban on AI in nuclear command or shared testing standards. Otherwise this is a photo op in which two leaders promise to be careful and tell nobody what that means. The public does not want vague assurances from the powerful. It wants rules it can enforce.


