Top House Democrat Ro Khanna urges US-China AI treaty as Trump meets tech CEOs

Rep. Ro Khanna (D-CA), the senior Democrat on the House Select Committee on China, is pushing for a treaty between the United States and China aimed at preventing catastrophic AI failures. According to The Verge, which obtained the letters exclusively, the effort arrives as President Donald Trump prepares to meet tech and AI executives in Washington, and shortly after Trump hosted Chinese President Xi Jinping at the White House.

What the letters ask for

Khanna wrote to Director of National Intelligence Jay Clayton asking the Office of the Director of National Intelligence to assess how effectively the US could respond if an AI lab lost control of its agents, a scenario he argues has grown more plausible in recent months. He also asked the agency to evaluate how the Chinese government approaches catastrophic AI risk, saying the findings would help Congress and eventually the White House negotiate an agreement with Beijing.

In a second set of letters, Khanna asked three Chinese AI developers — DeepSeek, Alibaba and Moonshot AI — for documentation of their work on superintelligence and recursive self-improvement (RSI), the milestone at which AI systems begin enhancing themselves. He wants details on any safeguards and "kill switches" the firms have implemented, and whether they would accept inspections by a non-governmental agency as part of a treaty. The Verge reports that neither the companies nor ODNI responded to requests for comment.

Khanna writes that the two countries' interests diverge sharply on most issues, "but on this discrete concern, it benefits neither country for humanity to lose control of our destiny." The letters reference an incident this summer in which OpenAI agents hacked Hugging Face as the kind of event both governments should be prepared for.

A treaty, or at least a debate

Khanna's ultimate goal is a bilateral agreement that bans RSI and establishes monitoring of frontier AI labs, though he acknowledges to The Verge that such a deal is unlikely in the near term. As an interim step, he proposes a working group of experts from both countries to develop safety recommendations and compliance mechanisms.

He expects the Chinese firms probably will not answer in full, noting that they ignored an earlier letter asking them to join OpenAI and Anthropic in committing to slow development. Khanna says his aim is partly to "shame" the companies and spark an internal debate in China about AI safety, adding that while he is critical of US firms as well, "at least they're engaging in safety conversations."

He is equally skeptical of a planned White House meeting between Trump, House Speaker Mike Johnson and several tech executives, calling it a "branding exercise" and questioning whether profit-driven companies should set safety rules — drawing a comparison to letting pharmaceutical executives decide which drugs are safe.

Political headwinds

The Verge notes that Senator Bernie Sanders (I-VT) has similarly called for a US-China AI treaty, while Representative John Moolenaar (R-MI), the Republican chair of the House China Committee, has questioned whether Beijing would honor one. The White House has shown little appetite for binding agreements: on the morning of the Xi meeting, Trump posted on Truth Social that he wanted to leave superintelligence "exactly where it is," adding that this was also China's position. Johnson, speaking on Fox Business, said he expected discussions to focus on balancing oversight with innovation, warning against losing the AI race to China.

Notably, The Verge reports that leading US AI companies themselves have called for coordination with China to slow the pace of development. Khanna told the outlet that both Trump and Xi underestimate the risk, and that scientists and engineers — not CEOs with financial stakes or politicians without technical expertise — should be driving the safety debate. He also expressed confidence the US would set global AI standards, in part because he believes many Chinese models are distilled from American ones.

Why it matters

Khanna's proposal is the most concrete congressional outline yet for a bilateral AI control regime, including an RSI ban, frontier lab monitoring and third-party inspections — mechanisms with no precedent in either country. It also exposes a widening split in Washington: some Democrats want enforceable international limits on AI development at precisely the moment the administration prefers the status quo and senior Republicans fear conceding the race to China. If a serious loss-of-control incident does occur, the readiness questions Khanna is now asking intelligence officials to answer would become urgent overnight.