Washington Draws a Chip Line, but Leaves AI Safety Talks Open

Avatar photo

ByRyan Mitchell

November 19, 2026

The Trump-Xi meeting highlighted two competing priorities: restricting China’s access to advanced U.S. chips while keeping room for cooperation on AI safety.

The Trump administration and Congress are trying to separate two questions often treated as one: whether China can obtain advanced American chips, and whether Washington and Beijing can communicate about artificial intelligence risks. At a meeting in Shenzhen on Thursday, President Donald Trump and Chinese President Xi Jinping addressed both, but their public statements produced no announced joint safety mechanism and no easing of chip restrictions.

Xi called for an action plan to build a “constructive China-U.S. relationship of strategic stability” and urged stronger communication on AI risks, including keeping AI under human control. Trump said the two sides should maintain positive momentum and that the United States does not support “Taiwan independence.” At the APEC summit, he also rejected global regulation of frontier AI, saying it could stifle the industry while acknowledging the technology needs careful management.

The practical distinction is significant. Dialogue about risks does not authorize technology transfers, and the leaders did not announce a deal to share advanced chips or establish binding global AI rules. Nor did they put an emergency channel for reporting serious AI incidents into operation. Reuters reported that a proposed notification system discussed by Treasury Secretary Scott Bessent and Chinese Vice Premier He Lifeng in September had not begun operating as of Nov. 16.

On technology controls, a U.S. interim rule targets a route around export restrictions: Chinese AI companies’ access to advanced U.S. chips through overseas data centers. CBS reported that the Treasury Department and National Security Council drafted the rule and Commerce issued it. The supplied reporting does not detail the rule’s full scope, enforcement provisions or implementation timetable. Those details will shape the compliance burden for cloud providers and companies seeking computing capacity outside China.

Congressional proposals could add further restrictions. OpenAI is backing bipartisan measures to limit future sales of high-end chips to China, counter diversion of Nvidia processors and restrict China’s access to sophisticated ASML equipment. These are proposals, not enacted law. Their effects depend on legislative text, passage and subsequent regulations; companies should distinguish them from controls already in force.

Rep. Lori Trahan, a Democrat who introduced a bill requiring oversight of advanced AI models, says chip controls and international safety cooperation need not be mutually exclusive. “Our most powerful chips landing in the hands of China is a different conversation than us collaborating globally around AI safety standards,” she told POLITICO. Her position reflects an opening for safety discussions even as lawmakers scrutinize technology transfers.

That opening does not amount to agreement on standards or a commitment to share sensitive systems. POLITICO reported that China-based safety group Concordia AI met with U.S. AI companies, including OpenAI, to discuss international collaboration and safety efforts in China. The discussions show industry-level contact, not a government agreement or shared enforcement regime.

For Nvidia and other firms, the immediate question is whether restrictions will limit sales, overseas computing access or both. Reuters reported that Nvidia’s current guidance includes no licensed Hopper products. However, the supplied material does not verify a reported sales figure or establish that the company’s position resulted from a new government order.

The next steps are procedural: agencies must clarify the interim rule, Congress must consider the chip proposals, and officials could still activate the unlaunched incident-notification system. Until then, Washington is pursuing a two-track approach—protecting advanced computing capacity while leaving a channel open for limited AI safety discussions.

Leave a Reply

Your email address will not be published. Required fields are marked *