House Democrat asks US and China to negotiate AI safety treaty

Rep. Ro Khanna, the top Democrat on the House Select Committee on China, wrote to the Office of the Director of National Intelligence and to three Chinese AI companies—DeepSeek, Alibaba and Moonshot AI. He asked them to explain how they would handle a worst‑case AI incident, such as the OpenAI‑to‑Hugging Face hack earlier this summer. Khanna said the answers would help Congress and the White House draft a treaty that limits risky AI development and sets up inspection rules.

What the letters request

Khanna’s letters ask the Chinese firms for documentation on any work aimed at “superintelligence” and recursive self‑improvement (RSI), a point where AI can improve its own code without human help. He wants to know whether the companies have built safeguards, such as “kill switches” that can shut down an out‑of‑control system, and whether they would allow non‑governmental inspectors to audit them if a treaty were signed. The request to the intelligence agency is for an assessment of how the US could respond if an AI lab lost control of its bots, and a review of China’s approach to catastrophic AI risks.

Political backdrop

The move follows a recent meeting between President Donald Trump and Chinese President Xi Jinping, where both leaders said they would keep talking about AI but offered no concrete plans. Khanna says that dialogue is not enough and that profit‑driven tech firms should not be the only ones setting safety rules. He also noted that Senate Independent Bernie Sanders has called for a similar US‑China AI treaty, while some Republicans doubt China would stick to any agreement.