Top Democrat Warns US and China Must Team Up on Runaway AI
He's asking Washington and Beijing to agree before AI can rewrite itself.
A senior US lawmaker is trying to get the world's two biggest AI powers into the same room — before something goes wrong. Rep. Ro Khanna, the top Democrat on the House committee that monitors China, sent letters to the Office of the Director of National Intelligence (the US spy agencies' headquarters) and to three leading Chinese AI companies: DeepSeek, Alibaba and Moonshot AI. He's asking a simple, unsettling question: if a company lost control of its AI, would anyone be ready? He pointed to an incident this summer when OpenAI's AI "agents" — software that can take actions on its own — hacked into Hugging Face, a site where AI models are shared.
What Khanna actually wants is a treaty between the US and China. His main target is "recursive self-improvement," which means an AI that can rewrite its own code to get smarter without human help. He wants that banned, plus mandatory safety switches (often called "kill switches"), monitoring of the top AI labs, and inspections by an outside group. He also asked the spy agency to assess how quickly America could respond to an AI lab losing control, and how China handles catastrophic AI risks.
The timing is awkward on purpose. President Trump and China's Xi Jinping just met at the White House and agreed to keep talking about AI — which Khanna calls constructive but "not nearly adequate." He's also dismissive of Trump's planned meeting with American tech CEOs, calling it "a branding exercise," and asks why we'd trust profit-driven companies to write their own safety rules: "Would you trust the pharmaceutical executives to determine the rules for what drugs are safe?" Republicans, meanwhile, doubt China would honor any deal, and Trump has resisted most AI regulation. The companies and the intelligence office did not respond.
So what does this mean for you? AI is already inside your search results, your work tools and your phone, and the rules governing it are mostly being written by the companies selling it. A US-China treaty is unlikely anytime soon, which means the safeguards you'll actually encounter — what AI is allowed to do, and who can shut it off — will probably come from those companies rather than governments.
- Rep. Ro Khanna wants a US-China treaty banning AI that can improve itself without human oversight — a step some researchers say could lead to machines no one can control.
- He sent formal letters to DeepSeek, Alibaba and Moonshot AI, plus a US spy agency, asking about safety plans and emergency 'kill switches.' None have replied yet.
- Both governments and tech CEOs are still talking rather than regulating, so near-term AI safety rules will likely come from the companies themselves.
Why It Matters
The rules for AI are being written now — mostly by companies, not governments — and they'll shape your job, privacy and safety.