US and Chinese officials plan to meet again in Shenzhen in about two months to discuss AI risks, Treasury Secretary Scott Bessent said Monday. He said the countries had formalized an AI dialogue and that Washington wants a direct communications line for serious incidents.
Speaking to CNBC after Sunday’s talks with Chinese Vice Premier He Lifeng in New York, Bessent identified uncontrollable AI agents, misuse by non-state actors, cyber threats and biological weapons as subjects for discussion.
He described the incident line as a channel officials want to establish. The two sides would also need to discuss protocols and agree on the main AI dangers.
Bessent also said AI companies must remain responsible for their systems. He rejected the idea of the government taking that responsibility away through a liability shield.
China’s official account confirmed that He, Bessent and US Trade Representative Jamieson Greer discussed AI during economic and trade consultations on September 20. It did not announce an incident line or a follow-up meeting in Shenzhen.
The Chinese statement also described discussions on carrying out earlier trade agreements and encouraging bilateral trade and investment. It characterized the exchanges as candid, in-depth and constructive.
Bessent’s proposal follows calls for a permanent channel connecting the countries’ AI experts. In a September 18 letter to the Treasury secretary, Americans for Responsible Innovation urged the governments to create an authenticated, continuously staffed communications link.
The advocacy group proposed modeling that channel on earlier arrangements for reducing risks from high-consequence technologies. It also called for a joint expert group to develop a shared glossary, so officials would use common terms when discussing AI risks.
The letter urged both countries to reaffirm their 2024 commitment to keep decisions on the use of nuclear weapons under human control. It proposed twice-yearly expert summits through 2028, with dates and reciprocal hosting arrangements agreed in advance.
The letter argued that AI-assisted cyberattacks, chemical or biological weapons development and accidental loss of control could threaten both countries. It warned that AI’s growing role in national-security systems could contribute to unintended escalation.
AI developers have reported cases of potentially dangerous use. Anthropic in September disclosed five biological-research cases it said could support weapons development. The company blocked the accounts but cautioned that the research could have legitimate purposes and did not establish harmful intent.
The letter, signed by the group’s president, Brad Carson, argued that initial cooperation could proceed without either country sharing sensitive capabilities or accepting binding commitments before trust and verification arrangements exist. It called for sustained discussions involving a consistent group of diplomats and technical experts.
Sources: CNBC, Xinhua, Americans for Responsible Innovation
–
By the Control Plane Editorial Team