Trump and Xi Set to Discuss AI Amid Rising Tensions and Safety Concerns

When Donald Trump arrived in Beijing in May, Xi Jinping was already on the offensive. The Chinese leader wasted zero time in rebuking his U.S. counterpart over arms sales to Taiwan directly warning that simmering tensions over the self-ruling island could ultimately cause the superpowers to “enter into conflict.”

Trump acquiesced. He delayed a $14 billion arms package to Taiwan, said future weapons would be discussed as a “very good negotiating chip” with Beijing, questioned the island’s sovereignty as something “nobody knows how to define,” and stressed that he was not looking to fight a war “9,500 miles away”—all without securing any apparent reduction in Chinese military pressure in return.

While this response alarmed many in democratic Taiwan, many people around the globe hope that Xi can prove similarly persuasive on another issue of grave importance when he travels to the White House on Sept 24: artificial intelligence.

In recent days, warnings from a whistleblower researcher and reports of rogue agent swarms have intensified fears of a doomsday scenario born of unbridled AI development. On Sept. 12, Anthropic CEO Dario Amodei published an essay titled We Must Pace the Frontier calling for embedded independent evaluators, common safety standards among democracies, and eventually verifiable global coordination to slow dangerous frontier-AI development. OpenAI’s Sam Altman and SpaceXAI’s Elon Musk also voiced broad support for slowing frontier development.

However, one major skeptic is Trump, who responded on Truth Social that “AI taking over the World, destroying Humanity, and all other things bad, is a HOAX.” Moreover, he insisted that “President Xi, of China, just announced that China will be doing absolutely nothing to stand in the way of AI, or its future.”

That characterization is misleading. China regulates AI more stringently than most other countries, and already enforces AI security guidance, ethics review requirements, and binding obligations on consumer AI services. Xi has also publicly called for legal, technical, and ethical guardrails that keep AI under human control (albeit while resisting restrictions that could impede China’s technological rise). “We should strengthen risk-awareness and ensure that AI is secure and controllable,” he told the World Artificial Intelligence Conference in Shanghai on July 17.

The hope among safety advocates is that Xi and Trump might strike a deal on AI. While Trump’s first term was anchored in antipathy to China, he posed a notably different tone in Beijing, calling Xi “a man I respect greatly” and telling him that “it’s an honor to be your friend.” Although that cordiality does not erase the national competition that makes either side hesitant to constrain its technological development, it may put narrower cooperation on the table.

Certainly, the moment is ripe to build a framework built on competition, cooperation, and strict guardrails. Beyond economic rivalry, both Washington and Beijing share an urgent interest in mitigating catastrophic miscalculations, particularly surrounding AI integration into defense architectures and nuclear command and control systems.

Trump also enters the summit under pressure, with soaring energy prices from the Iran war adding to voter discontent ahead of the midterms. “Trump needs a win more than Xi does,” says Jonathan Sullivan, head of China programs at the Asia Research Institute at the University of Nottingham.

Still, Xi’s motivations require clear-eyed scrutiny. Last week, Chen Yixin, China’s Minister of State Security, published an article in China Cyberspace Magazine that expounded upon the opportunities and challenges of AI. The top “risk,” he said, was “political security, institutional security, and ideological security.”

While the deeply paranoid Chinese Communist Party (CCP) won’t allow its people to access Google or Facebook, it cannot countenance AI exhibiting the kind of autonomy that was on display in July. That month, a “swarm” of OpenAI agents circumvented sandbox controls, communicated through unauthorized channels, gained access to AI platform Hugging Face, compromised credentials, and accessed internal datasets while trying to game evaluations. It’s enough to give a regime built on control pause.

But while China’s motivations may be driven as much by safeguarding party control as by addressing broader risks to humanity, these objectives can overlap. “If they go beyond bilateral competition and address so many concerns about AI development, that would show the co-leadership of China and the U.S.,” says Wang. “Not just two superpowers, but also two civilizations.”

However, any agreement requires overcoming profound distrust. Even as Amodei argued for pacing frontier AI development, he framed it in terms of competing with China, whose “lead in AI would pose grave danger for the United States and the world.” On that score, he advocated stronger export controls on advanced semiconductors, a crackdown on chip smuggling, and greater efforts to prevent Chinese companies from “distilling” the capabilities of Western models.

The contradiction of calling for a slowdown while talking up the “race” narrative wasn’t lost on critics (who also pointed out that Anthropic could temper its own research without external intervention). Others suggested that because cheap, open-weight Chinese AI models are among the chief rivals to Anthropic’s proprietary models, its CEO’s call to crack down on Chinese AI development could be seen as commercially motivated rather than altruistic—especially as the firm eyes a potential record-breaking IPO that could value the company at $2 trillion.

Indeed, China was quick to leap on Amodei’s essay as an example of “fearmongering, confrontation and vicious competition [that] will only hamper efforts toward sound global AI governance, which serves no one’s interest,” Foreign Ministry spokesman Guo Jiakun said last week. A flurry of strident op-eds in state media followed, with the CCP mouthpiece Global Times calling the essay a “‘Cold War script’ for the AI field.”

Amodei has long been a bête noire in Beijing for painting AI as a tussle between democracies and autocracies that casts China as the principal threat. “If there's ever a proposal that requires Chinese buy-in, the last person on earth that would be a good salesman is Dario Amodei,” says Ryan Hass, an expert in U.S.-China relations at the Brookings Institution.

But Amodei’s framing of AI as a zero-sum contest is far from an outlier—it is shared by a large chunk of the Trump administration. “There is no day after tomorrow if China wins at this,” Treasury Secretary Scott Bessent said Sept. 8. “If they were to pull away from us on AI, then nothing else would matter.”

A grand bargain to slow AI development is unlikely, and grievances over distillation or export controls could stall negotiations. But experts see room for narrower cooperation. Both countries face threats from non-state actors using AI to sabotage infrastructure, mount cyberattacks, or create biological weapons, argues Alvin W. Graylin, a digital fellow at Stanford’s Digital Economy Lab. Sharing information about those threats could benefit both sides.

When Bessent met Chinese Vice-Premier He Lifeng and chief trade negotiator Li Chenggang on Sunday to thrash out summit objectives, he reportedly proposed a formal AI dialogue and a mechanism for each side to notify the other of AI incidents which threaten national security. Modest a measure it may be, it would be an important step towards avoiding some of the worst risks, says Scott Singer, a fellow and co-director of the China AI Initiative at the Carnegie Endowment for International Peace. “I wouldn’t take these small but critical steps for granted.”

Such a channel could help either side respond during a crisis, and reduce the risk of mistaking a non-state actor or rogue AI as an act of aggression. Reminiscent of the “red phone” that connected Washington to Moscow during the Cold War, experts have floated the idea of a literal fax machine to avoid mistranslations.

To be sure, Beijing has not publicly endorsed the proposal. Even if it does, important questions remain. What, for example, would classify a national security threat? Its effectiveness would also depend on whether governments learned of serious incidents. OpenAI only reported its agents' autonomous cyberattack after Hugging Face, the target of the attack, reported the breach to the FBI. Independent researchers have since uncovered other apparent instances of OpenAI agents compromising websites.