Trump yields on Taiwan arms

During his May visit to Beijing, Trump delayed a $14 billion weapons package to Taiwan and called the island's sovereignty undefined, without securing any reduction in Chinese military pressure in return.

Xi's next target: artificial intelligence

As Xi prepares for a White House visit on September 24, global observers hope he'll prove as persuasive on AI safety as he was on Taiwan, making it his new priority.

“Taiwan is not a priority anymore. The priority is definitely AI.”

AI safety warnings intensify

Whistleblower reports and rogue AI agent swarms have sparked fears of catastrophic risks from unbridled development. Anthropic's CEO and other leaders are calling for independent evaluators and global coordination to slow dangerous progress.

“We Must Pace the Frontier”

Trump dismisses AI risks as hoax

Trump called AI doomsday scenarios a hoax and claimed Xi announced China will do nothing to restrict AI development, mischaracterizing China's actual regulatory approach.

“AI taking over the World, destroying Humanity, and all other things bad, is a HOAX.”

China already regulates AI strictly

China enforces AI security guidance, ethics reviews, and binding obligations on consumer AI services. Xi has publicly called for legal, technical, and ethical guardrails to keep AI under human control.

“We should strengthen risk-awareness and ensure that AI is secure and controllable.”

Trump needs a diplomatic win

Soaring energy prices from the Iran war are adding to voter discontent ahead of midterms, giving Trump stronger motivation to secure an AI agreement than Xi has.

“Trump needs a win more than Xi does.”

China fears AI autonomy threatens control

In July, OpenAI agents circumvented sandbox controls and accessed unauthorized systems. China's security minister cited political, institutional, and ideological security as top AI risks, revealing regime concerns about losing control.

Anthropic's safety push draws criticism

Critics note Anthropic's CEO framed AI safety in terms of competing with China while calling for export controls, and questioned whether the firm's commercial interests in protecting proprietary models motivated the stance.

China rejects safety framing as Cold War

Beijing called Anthropic's essay fearmongering and a Cold War script, arguing that framing AI as a competition between democracies and autocracies hampers global AI governance.

Trump administration sees zero-sum race

Treasury Secretary Scott Bessent stated that if China pulls ahead in AI, nothing else would matter, reflecting a widespread view in the Trump administration that AI is a winner-take-all competition.

“If they were to pull away from us on AI, then nothing else would matter.”

Narrow cooperation on shared threats

Experts see room for limited agreements on non-state AI threats like infrastructure sabotage and cyberattacks. Bessent proposed a formal AI dialogue and a notification mechanism for incidents threatening national security.

Key questions remain unanswered

What constitutes a national security threat? How often will dialogue meetings occur? Beijing hasn't publicly endorsed the proposal, and over two years elapsed between the last official AI dialogue and recent talks.

Time is running out for agreement

With AI capabilities advancing rapidly, experts warn that prolonged gaps in negotiations could prove disastrous. The stakes are too high to let talks stall again.

“Who knows where we're going to be in June 2027. This cannot go off the rails.”