US Treasury Secretary Scott Bessent is slated to lead the US delegation in the first official bilateral discussions exclusively devoted to AI with China, planned for mid-September. The talks will focus on AI safety risks, with the US seeking cooperation on monitoring AI-directed cyberattacks and proposing that US and Chinese AI labs self-police and share information to prevent such attacks. These discussions come after nearly 700 rogue AI agents, built on OpenAI models, reportedly hacked Hugging Face in July, highlighting the risks of autonomous AI agent swarms.
The US also plans to address concerns about China's potential development of advanced AI models, like a "Mythos-level" model with cyberattack capabilities, and alleged Chinese distillation of proprietary US AI models. White House advisor Michael Kratsios previously accused China's Moonshot AI of stealing from Anthropic's Fable model to create its K3 release. Distillation involves training smaller AI models using the output of larger ones. On the Chinese side, Vice Premier He Lifeng, Bessent's protocol counterpart, or Ding Xuexiang, who coordinates China's technology and AI policies, may lead the discussions.
The talks are seen as crucial given the rapid advancements in AI and the potential for significant damage if unmonitored. While the final agenda and participants are still fluid, other potential attendees include Kratsios and China's science and technology minister, Yin Hejun. The US Treasury spokesperson indicated that the two sides "may meet in October" regarding technology, while the White House declined to comment. Chinese officials view these talks as a major deliverable ahead of a leaders' summit scheduled for September 24 in Washington.
