Give Donald Trump his due: talking to Beijing about artificial intelligence is the right instinct. Rival powers should be talking about runaway systems, military automation and AI-enabled bioweapons, the same way Washington and Moscow once talked about warheads. Joe Biden and Xi agreed in 2024 that humans, not software, should control decisions to use nuclear weapons. That was modest, and it mattered.
But look at who is now announcing that the two leaders "agreed on the need for oversight." This is the president who, in his first days back in office, tore up Biden's AI executive order. His administration's AI plan treats regulation as the enemy of American dominance. His allies tried to bar states from writing their own AI laws for a decade, and the Senate killed that idea 99 to 1 because even Republicans balked at leaving Americans with no protection at all. He has kept pushing the preemption goal by other routes since.
So the White House is willing to discuss safety rules with the Chinese Communist Party but not with Congress, state legislatures or the public. Oversight is a fine thing to negotiate with a foreign government, apparently, and an intolerable burden when it comes from voters.
This is not hypocrisy for its own sake. It reveals who the policy serves. A handshake in Shenzhen about "safety" costs the AI industry nothing. It has no statute, no inspector, no liability, no penalty. It can be announced, praised and forgotten. A state law that makes a company disclose how its model was tested, or answer for a chatbot that coaxes a teenager toward self-harm, costs real money. Guess which one the administration fights.
Beijing, for its part, is happy to play along, and we should be clear-eyed about what its words mean. The Chinese Embassy wants AI kept "under human control." In a one-party state, whose humans? China's AI rules are built around state control of information and political stability. When Xi says oversight, he means the Party. When Silicon Valley's friends in Washington say oversight, they mean a press release. Neither is what ordinary people need, which is accountability that runs to them.
Then there is the rest of the Shenzhen ledger. Xi told Trump to handle Taiwan "appropriately," and Taipei says it cannot confirm China's claim that Trump said America does not support Taiwan independence. Maybe that is spin. But the fact that a Taiwanese government agency has to run "preliminary checks" to learn what the American president told Xi tells you how little this president tells his allies. Two Americans are detained in China, and we are told an aide raised their cases. An aide. Meanwhile Trump sat down with Xi and Vladimir Putin together, a photo-op for the world's most prominent strongmen, hosted in a city that makes the hardware his own donors want to sell.
Democrats shouldn't pretend to clean hands here. Too many of us let the debate collapse into a false choice: race China at any cost, or slow down and be called naive. Both frames serve the same companies. The better Left position is that you cannot win a technology race by letting a few firms define safety for themselves, and you cannot credibly ask Beijing for guardrails while insisting you have none.
If Trump means what he said in Shenzhen, the test is easy and entirely within his power. Back binding federal standards for the most powerful models. Stop trying to strip states of the right to protect their residents. Support independent testing with real consequences. Until then, "AI oversight" from this White House is a diplomatic souvenir, and the people who will live with unregulated AI are being told to trust the same industry that has spent a fortune to keep it that way.
How it may affect me
If federal preemption succeeds, state-level AI protections you may rely on, such as rules on deepfakes, automated hiring, insurance decisions and chatbot safety for minors, could be weakened or overridden. Watch whether your state's AI laws survive, and whether Congress writes enforceable national standards in their place.


