China frames AI safety as protecting Communist Party rule
President Trump and Chinese President Xi Jinping plan to discuss artificial intelligence at a Washington summit next Thursday, with U.S. Treasury Secretary Scott Bessent — who is leading the U.S. artificial-intelligence track — set to meet He Lifeng, Xi’s top economic official, in New York this weekend to prepare, according to people familiar with the matter. The meeting will be Xi’s first state visit to Washington since 2015.
The two governments view AI safety very differently, the Wall Street Journal reported. In the United States, the discussion has centered on guarding against a powerful model that slips beyond human control, an autonomous weapon that acts on its own, or a tool potent enough to help a rogue actor build a weapon. Trump has played down the need for regulatory guardrails, the Journal reported, though prominent American AI executives have called for a slowdown in development to weigh safety first.
In China, AI safety increasingly points somewhere else — at threats to the political order the Communist Party sits atop. The gap, the Journal wrote, makes a durable dialogue — the kind that might keep the competition from spiraling into a crisis neither government is prepared to manage — a tall order.
China’s priorities were on display this week when Chen Yixin, China’s minister of state security, laid out the government’s view of AI risk in a party publication. The first danger he named was not to humanity but to the state. AI, he warned, could “systemically affect the political-security environment,” giving hostile forces the means to mass-produce rumors, wage “cognitive warfare” and stir unrest. His prescription was not technical guardrails but tighter party control, including upholding “party management of the internet and party management of data.”
To some in Washington, that mismatch is unlikely to be negotiated away. Chris McGuire, who worked on technology and China policy at the National Security Council under President Joe Biden and is now a senior fellow at the Council on Foreign Relations, said Beijing “has consistently rejected overtures of serious collaboration on AI safety with the U.S.” China’s “number-one priority is becoming the global leader in AI, and it’s unlikely to implement measures that would slow down its AI development as long as its own AI capabilities are mere months behind the leading U.S. labs,” McGuire said.
Trump and Xi committed to some form of AI dialogue at their Beijing meeting in May, though little has happened ahead of the Chinese leader’s arrival next week. Plenty else will crowd the agenda when the two leaders meet next Thursday — Taiwan, Iran and trade — along with the pomp of a formal state dinner. But AI looms large, the Journal reported.
In a statement to the Journal, Bessent said: “the United States remains the leader in AI. We are open to discussions on avoiding shared risks and avoiding bifurcation of our two systems. We expect our discussions to cover both open and closed weight models.” America’s leading labs, such as OpenAI and Anthropic, keep their most capable models closed, reachable only through paid gateways, while China’s top players, including DeepSeek, release theirs as open-weight systems free for anyone to download — a bet, analysts said, that giving the technology away will spread Chinese AI worldwide and blunt U.S. efforts to contain it.
The mismatch in priorities is not new, and the last attempt at hashing out differences over AI is a cautionary tale. When the Biden administration opened a formal AI dialogue with Xi in 2023, Beijing put its foreign ministry in charge rather than a technical body, an arrangement that limited the substance of the exchange, according to U.S. officials involved in those talks at the time.
Some policymakers said the U.S.-China competition over AI resembles the Cold War dynamic between the United States and the Soviet Union. Robert Hormats, who served on the National Security Council under Henry Kissinger, said the two superpowers had reconciled themselves to continuing to build bigger and better weapons, but also thought it was possible and urgently necessary, at the same time, to agree on “a set of rules and a common effort to avoid mutual catastrophe.” The question now, he said, is whether the U.S. and China can identify common areas of interest, establish procedures for mutual risk avoidance and begin sustained high-level negotiations with seasoned experts to build on those. The U.S. hasn’t resolved that argument at home — which makes it harder to press one abroad.
China’s own signals are mixed. When American business and think tank representatives met in Beijing last month with Wang Huning, the Communist Party’s leading ideologue and one of its most senior figures, he appeared to agree on the importance of a bilateral discussion on AI guardrails, said people familiar with the matter.
Others see a blind spot. Gary Rieschel, founding partner of Qiming Venture Partners, said his impression from meeting recently with Chinese officials in Beijing is that “they are underestimating the threat they face from AI intrusion, because they think they can fend it off with the Great Firewall.” The Great Firewall is a catchall term for China’s decadeslong effort to seal the country’s internet off from the outside world, blocking foreign platforms and scrubbing unwelcome content for a billion-plus users. One of the most sophisticated systems of information control ever built, it is a pillar of the party’s hold on power. The Firewall is built to control what Chinese users see, not to repel intrusions into networks and systems. An AI that can probe for weaknesses on its own, analysts said, is a different kind of threat than a foreign website or a social-media post — one a wall cannot keep out.
For all their confidence, however, Chinese officials have been eager to know where the AI talks stand — and whether the dialogue might help ease U.S. restrictions on high-tech sales to China, the Journal reported. Chris Slevin, who visited Beijing in July as a commissioner of the U.S.-China Economic and Security Review Commission, said Chinese officials peppered his delegation with questions about the bilateral AI dialogue. In one meeting with foreign-ministry officials, roughly half of the hour was spent on AI and related topics, Slevin said; the Chinese side asked how the U.S. was organizing the talks and whether chip-export controls would be on the agenda. The commission’s requests to visit major Chinese AI labs, including DeepSeek and Moonshot, were denied.
Slevin, who also discussed AI with Chinese officials three years ago as a Commerce Department official, said they were “more self-assured” this time about the state of their technology, stressing that the two countries had reached rough parity and telling their American guests that Chinese AI models are open for the world to use.
An early sign of how serious Beijing will be about the AI talks will be whom it designates to lead such a dialogue. But naming the people, like announcing the dialogue itself, is the easy part. As the Journal wrote, what no communiqué can resolve is that one side wants to make the technology safe, and the other wants to make it safe for the party.