A House Democrat seeks a US-China treaty on self-improving AI

Representative Ro Khanna is calling for a US-China agreement that would ban recursively self-improving AI and monitor frontier laboratories. As a first step, he asked US intelligence officials and Chinese AI companies to explain how they would audit and respond to a loss-of-control incident.

Letters to DeepSeek, Alibaba and Moonshot AI request information about superintelligence work, safeguards and kill switches, plus whether the companies would accept inspections by a nongovernmental body. Khanna separately asked the Office of the Director of National Intelligence to assess US response capacity and China’s approach to catastrophic AI risks. None had responded when The Verge published its report.

Khanna acknowledges that a treaty is unlikely soon and proposes a bilateral expert group as an interim measure. The plan faces skepticism over verification and whether either government would constrain domestic labs during strategic competition. For now, the letters begin an information-gathering effort rather than negotiations, and they do not create obligations for the companies contacted.