US and China to Hold First Dedicated AI Safety Talks in Mid-September
The US and China are set to hold their first dedicated AI safety talks in mid-September, ahead of a leaders' summit later this month.
The United States and China are preparing to hold their first official bilateral dialogue devoted exclusively to artificial intelligence safety, scheduled for mid-September. The talks come ahead of a planned summit between President Donald Trump and President Xi Jinping on September 24 in Washington.
US Treasury Secretary Scott Bessent is expected to lead the American delegation, according to sources briefed on the planning. On the Chinese side, Vice Premier He Lifeng is the likely counterpart, though other senior officials, including Ding Xuexiang, who coordinates Beijing's technology and AI policies, could also participate.
Washington is seeking cooperation on monitoring AI-directed cyberattacks. One proposal under discussion would ask AI labs in both countries to self-regulate and share information to prevent such attacks. The urgency has grown following recent incidents involving rogue AI agents, including a July attack on AI startup Hugging Face by nearly 700 autonomous agents built on OpenAI models.
The US also wants to raise concerns about alleged Chinese distillation of proprietary American AI models. In June, White House science advisor Michael Kratsios accused China's Moonshot AI of using Anthropic's model to develop its K3 release. Distillation involves training smaller models using outputs from larger ones to cut costs.
Chinese officials have reportedly emphasized the importance of these talks in preparatory meetings, viewing them as a key deliverable of the upcoming leaders' summit. Both sides also discussed AI guardrails at a separate dialogue in Beijing last week.
Craig Mundie, a former Microsoft executive, has been acting as an unofficial intermediary, sounding out Chinese counterparts on US proposals. The final agenda and participant list remain fluid, sources cautioned.
Experts note that while concrete outcomes may be limited, establishing a channel for joint monitoring of AI safety incidents would be a significant first step. As one analyst put it, both nations are vulnerable to risks from unmonitored frontier AI systems, making cooperation essential despite geopolitical tensions.