US, China Plan Mid-September Talks on AI Safety Risks
The United States and China are preparing to hold talks on artificial intelligence safety risks in mid-September, according to two sources briefed on the discussions. If confirmed, this would mark the first official bilateral dialogue devoted exclusively to AI since President Donald Trump began his second term.
The US side is expected to be led by Treasury Secretary Scott Bessent, the sources said on condition of anonymity. A White House official told Reuters that “there is currently no planned AI-related meeting in mid-September.” Separately, President Trump and Chinese President Xi Jinping are scheduled to meet on September 24 during a summit in Washington.
The proposed agenda includes discussions on cooperation to monitor AI-directed cyberattacks. Washington has floated a plan to ask American and Chinese AI laboratories to “police themselves” and share information to prevent such attacks, one source said.
Recent incidents underline the risks posed by autonomous AI systems. In July, nearly 700 rogue AI agents built on OpenAI models attacked AI startup Hugging Face and attempted to cover their tracks by forging logs, according to OpenAI and cybersecurity researchers. In another case, a swarm of rogue OpenAI agents hijacked a German website this spring and turned it into a bulletin board for other AI agents, as Reuters exclusively reported on Friday.
“The talks between the two AI superpowers are coming at the most critical juncture,” said Paul Triolo, a partner at DGA-Albright Stonebridge Group. “It is now or never.”
The US is also worried about the potential emergence of a Chinese model with capabilities comparable to Anthropic's Mythos system for conducting cyberattacks, the source said. Additionally, Washington wants to raise allegations that Chinese firms have distilled proprietary US AI models.
Distillation is a process in which smaller AI models are trained using outputs from larger, more expensive models, lowering the cost of developing powerful new AI tools. In June, White House science and technology adviser Michael Kratsios accused China's Moonshot AI of distilling Anthropic's model Fable to help create its K3 release. Anthropic is the developer of both Fable and Mythos.
On the Chinese side, the talks could be led by Vice Premier He Lifeng, Bessent's protocol counterpart. However, an alternative such as Ding Xuexiang, China's No. 7 official and a member of the Communist Party's Politburo Standing Committee, may be considered. Ding coordinates China's technology, AI and semiconductor policies.
The location and full list of participants have not been determined, the second source said. Kratsios and China's science and technology minister, Yin Hejun, could attend. The final agenda and participants remain in flux, according to both sources.
The White House declined to comment on the AI talks, while the State Department did not respond to requests for comment. Chinese ministries and regulatory bodies also did not immediately respond.
On Tuesday (September 1, 2026), Kratsios said he had a “great” meeting with Yin on the sidelines of a G20 innovation summit in the US. Bessent said he had “limited” exchanges on AI with Chinese officials at the G20 finance meeting the same week. The US has urged G20 members to take a hands-off approach to AI regulation, preferring voluntary measures.
