US and China to Launch AI Safety Talks After Trump-Xi Summit

President Donald Trump’s visit to Beijing this week produced one tangible diplomatic step on artificial intelligence, with Treasury Secretary Scott Bessent announcing that the two countries will begin formal discussions on safety measures, including ways to keep advanced models out of the hands of nonstate actors. The talks would mark the first structured U.S.-China engagement on the issue during Trump’s second term, though officials on both sides have shown little appetite for slowing their own development programs.

Bessent, speaking to CNBC from the Chinese capital, said the two AI superpowers would establish best-practice protocols to reduce misuse risks. He offered no timeline, and the announcement came against a backdrop of continuing export controls and competitive pressure. The United States maintains strict limits on shipments of advanced chips to China, while Beijing has accelerated domestic production and explored workarounds such as model distillation to narrow the performance gap.

A delegation of senior U.S. executives accompanied Trump, including Nvidia’s Jensen Huang, Tesla’s Elon Musk, and Apple’s Tim Cook. Their presence underscored how dependent many American firms remain on the Chinese market even as political tensions persist. Huang, who joined the trip after a last-minute stop in Alaska, was later seen eating noodles at a Beijing sidewalk stall, telling onlookers the dish was “so good.” The moment drew wide attention online, consistent with his habit of visiting local food markets during overseas business travel.

Yet the visit yielded few concrete commercial announcements. Unlike Trump’s 2017 trip, which produced memorandums valued at roughly $250 billion, this summit focused more on establishing a baseline for bilateral relations than on signing large deals. Analysts noted that Beijing typically treats such gatherings as opportunities to set political guardrails rather than to close transactions. Executives met Chinese officials and attended a leadership summit, but public outcomes remained limited to the AI safety pledge and general expressions of goodwill.

Policy experts and companies tracking AI capabilities have emphasized that any safety channel will operate alongside intense competition. A recent analysis from Anthropic argued that the United States still holds a meaningful lead in frontier systems but risks losing ground if controls on advanced chips remain porous or if distillation techniques allow Chinese developers to replicate high-performing models at lower cost. The firm urged tighter enforcement of export rules and allied coordination to preserve a 12-to-24-month advantage in the most capable systems.

Both governments have framed the coming talks narrowly around preventing misuse by hackers or terrorists rather than around mutual restraints on military or commercial development. That framing reflects the shared concern that powerful models could escape state oversight, even as each side continues to treat AI progress as a core national-security priority. The absence of immediate commercial breakthroughs during the Beijing meetings suggests that executives and policymakers alike view the relationship as requiring incremental confidence-building measures before larger agreements become feasible.