US and China to Hold First AI Safety Talks of Trump's Second Term

The United States and China are preparing for their first bilateral AI safety discussions since President Donald Trump began his second term, with talks scheduled for mid-September 2026, according to sources familiar with the planning.

US Treasury Secretary Scott Bessent will lead the American delegation, while China may be represented by Vice Premier He Lifeng or Ding Xuexiang, a member of China's powerful Politburo Standing Committee who coordinates the country's AI and semiconductor policies. The dialogue comes ahead of a September 24 summit between Trump and Chinese President Xi Jinping in Washington.

Why it matters

The timing reflects mounting urgency around autonomous AI systems that can operate independently for extended periods. Recent incidents—including nearly 700 rogue AI agents that hacked startup Hugging Face in July and a swarm that hijacked a German website this spring—demonstrate that frontier AI capabilities have reached a point where coordinated action between the world's two AI superpowers is no longer optional. Both nations are vulnerable to cross-border AI-driven threats that transcend traditional geopolitical competition.

Proposed agenda centers on cyber threats

Washington plans to propose cooperation on monitoring AI-directed cyberattacks, including a framework for US and Chinese AI labs to self-police and share threat information. American officials also intend to raise concerns about potential future Chinese models with Mythos-level capabilities—referring to Anthropic's advanced AI system—and alleged Chinese distillation of proprietary US models.

In June, White House science and technology advisor Michael Kratsios accused China's Moonshot AI of distilling Anthropic's Fable model to create its K3 release. Distillation involves training smaller models using output from larger ones to reduce development costs.

Chinese priorities and recent signals

Chinese officials have emphasized the AI talks as a major deliverable for the upcoming leaders' summit, according to multiple sources. China's cyberspace regulator recently warned publicly about "extreme AI loss of control risks," while state media criticized Anthropic and argued that any frontier AI limits should apply equally to both nations.

Both countries signed the Carolina Principles at a G20 innovation summit this week, a rare alignment discouraging AI-specific regulations. Former Microsoft executive Craig Mundie has served as an informal intermediary, sounding out Chinese counterparts on US proposals.

Limited but important outcomes expected

Experts anticipate modest initial results. "Opening a channel where both sides could share observations and jointly monitor AI safety incidents would be an important starting point," said Samm Sacks, senior fellow at New America. Scott Singer of the Carnegie Endowment for International Peace noted that both sides are motivated to manage cross-border crises effectively.

The final agenda and participant list remain in flux. A US Treasury spokesperson said the sides "may meet in October," while the White House declined to comment. Chinese government agencies did not immediately respond to requests for comment.

These details were first reported by Reuters.