Trump-Xi Summit Unlikely to Yield Major AI Safety Agreement

Limited expectations for U.S.-China AI talks

When President Donald Trump hosts Chinese leader Xi Jinping in Washington on September 24, artificial intelligence will be on the agenda — but don't expect a breakthrough on safety guardrails. According to Dr. Lily McElwee, president and CEO of the Phoenix Committee on Foreign Relations, deep mistrust between the two superpowers will prevent any sweeping agreement on managing AI risks.

"I expect very limited progress in these talks," McElwee said, speaking on the Generation AI podcast. The summit follows a week of warnings from AI industry leaders about the technology's rapid advancement and marks the second Trump-Xi meeting this year, after their May summit in Beijing.

Treasury Secretary Scott Bessent is meeting with Chinese Vice Premier He Lifeng ahead of the summit to discuss topics including AI, McElwee noted, though trade and other issues will also dominate the conversation.

Why it matters

As AI systems grow more powerful, the absence of U.S.-China coordination creates genuine risks. Without cooperation between the world's two AI superpowers, efforts to prevent dangerous applications — from autonomous weapons to nuclear command integration — remain incomplete. The trust deficit that prevents broader agreements also makes it harder to verify compliance with any narrow deals that do emerge.

China's domestic AI priorities differ from U.S. concerns

Beijing does recognize AI dangers, McElwee said, pointing to Xi's July statement that AI must remain under human control. But China's domestic regulations focus primarily on political risks, requiring training data to conform to Chinese values and ideological controls tied to Communist Party concerns about regime security.

"I do think that behind the scenes, Chinese leaders recognize some of these broader existential risks," McElwee said. "But that doesn't mean they're willing to have those conversations with U.S. leaders."

Chip controls remain non-negotiable

A recent essay by Anthropic CEO Dario Amodei highlighted the core tension: slowing AI globally requires China's cooperation, yet Amodei also called for maintaining restrictions on China's access to advanced chips. Beijing dismissed warnings from U.S. tech leaders as "fearmongering."

McElwee said Beijing interprets these warnings as evidence that China is keeping pace in frontier AI development, particularly after releases like the Kimi K3 model earlier this year. Beijing's chief concern is chip export controls, which the U.S. has made clear are not up for negotiation.

"So I see limited progress feasible here," McElwee said.

What narrow agreements might emerge

McElwee identified two outcomes as "tentatively achievable." First, a government-to-government channel or hotline to discuss AI risks and early safeguards. Trump and Xi agreed to such a channel in May, she said, but implementation has stalled.

Second, both sides might define specific AI behaviors they consider dangerous, such as removing human control from weapons systems or integrating AI into nuclear command chains.

A sweeping arms control agreement remains out of reach, McElwee said, largely because of a trust deficit that has deepened over 15 years. U.S. policymakers worry any deal could benefit China without real gains for America, given enforcement challenges.

The Atlantic Council raised similar doubts this week, pointing to a 2015 agreement between Xi and then-President Barack Obama to curb cyber theft of intellectual property. Three years later, U.S. trade officials concluded China had not fundamentally changed its practices. The think tank also questioned whether an AI crisis hotline would work, noting that Beijing reportedly turned down a U.S. request for a call between defense chiefs after the 2023 Chinese surveillance balloon incident.

Trump's stance and Arizona implications

President Trump has pushed back hard on calls to slow AI, calling fears of existential risk a "hoax." McElwee said the president sets the framework through the administration's AI strategy, which emphasizes global dominance and competition with China. Still, Bessent's openness to discussing broader risks shows "a little bit of nuance" that could lay groundwork for future talks.

For Arizona's growing semiconductor industry, including Taiwan Semiconductor Manufacturing Co., the biggest risk would be changes to chip export controls, McElwee said. She called that unlikely, meaning TSMC and the state's broader chip manufacturing sector should see limited impact from the summit.

These details were first reported by AZ Family.