U.S. and China Confirm AI Safety Talks to Be Held in Shenzhen in Two Months
nashnova research
Treasury Secretary Bessent confirmed the U.S. and China will meet in Shenzhen in roughly two months for dedicated AI safety talks — covering risk assessment, an incident hotline, and developer liability — the first time both sides aim to move from principles to an actionable agreement framework.
What will the two sides actually negotiate?
Both sides agreed to establish a formalized AI dialogue mechanism with three core topics: AI risk assessment, communication protocols for AI safety incidents, and a dedicated "incident hotline."
This means → U.S.-China AI engagement is shifting from ad-hoc meetings to a structured channel with a fixed agenda and an emergency line.
Bessent added that both sides want to build consensus on major AI risks — including rogue AI agents, non-state actors, and non-state actors in the cyber domain.
Why did Bessent call out OpenAI by name?
Citing the attack on the Hugging Face platform, Bessent said directly: "This is OpenAI management's responsibility, not a bunch of agents' responsibility."
He reiterated that AI developers should bear legal liability for their products — including scenarios where AI agents go rogue and launch uncontrollable cyberattacks.
In plain terms = Bessent's position is unambiguous — when AI causes harm, the lab pays the bill, not some defense of "the AI acted on its own." His exact words: "These labs can slow down at any time."
Will the tariff truce be extended?
The U.S.-China tariff truce is set to expire on November 10. Bessent confirmed the topic was discussed but gave no clear answer on whether both sides agreed to extend it.
He struck an overall positive tone on trade, saying the relationship has been "highly stable since last fall."
This means → No signal of a breakdown on tariffs, but no firm commitment either — ambiguity itself is the current negotiation tactic.
What is the real test for the Shenzhen meeting?
Until now, U.S.-China AI safety engagement has stayed at the level of statements of principle. The Shenzhen talks are the first explicit push toward an actionable agreement framework.
This reflects a larger signal: as the AI capability race accelerates, the world's two biggest AI powers are starting to seriously discuss what happens when something goes wrong.
In plain terms = Before, it was "we both agree AI safety matters." Now the task is to sit down and define how the hotline works, how liability is assigned, and how risks are categorized — moving from slogans to an operating manual.
市场有风险，内容仅供研究参考，不构成投资建议。