The US–China AI Line: a Smoke Detector, Not a Building Code

The US has proposed a “notification mechanism” with China for artificial intelligence incidents that could affect national security ahead of today’s Trump–Xi summit, Treasury Secretary Scott Bessent said Sunday.

The mechanism is essentially an open line between Bessent and Chinese Vice Premier He Lifeng, and Beijing has not publicly agreed, Politico reported, citing four people familiar.

“We want a shared vision of common goals and common threats.”

Scott Bessent, US Treasury Secretary, to reporters Sunday, as reported by AP

Detection without restraint

A smoke detector reports the fire. It doesn’t point to who struck the matches.

As described, the channel Bessent proposed works the same way: it would tell each side something has gone wrong, not limit what either builds. President Donald Trump has resisted calls to slow AI development, saying that would help China catch up to US companies. Both governments remain wary of measures that could slow their own technological development, Reuters reported. At a G20 innovation summit in the US this month, China signed the Carolina Principles, an agreement aimed at discouraging AI-specific regulation, while the US urged members to avoid new AI rules.

“…whoever wins AI, wins.”

President Donald Trump, to reporters during a recent trip to Ireland, per Politico

Export controls, the US limits on sales of advanced chips and chipmaking technology to China, stayed out of the room. Greer said export controls on AI chips were not part of the discussions. One source told Reuters on Sept 4 that Washington also wanted to raise alleged Chinese distillation of US models, meaning training smaller models on the output of larger ones, and its concern about a future Chinese model at the level of Mythos, from AI developer Anthropic, with the capability to conduct cyberattacks. None of the accounts of Sunday’s talks reviewed for this story mentions either.

What Beijing has Said, and Hasn’t

Bessent said on CNBC that the two countries had formalized the “USA-China AI dialogues” and agreed to meet in roughly two months in Shenzhen, per Politico. “I think the important thing was to start talking,” he said.

Beijing hasn’t said so.

Bessent and Greer met Chinese Vice Premier He Lifeng in New York on Sunday. Xinhua, China’s state-run news agency, described candid, in-depth and constructive exchanges that included a dialogue on AI, without specifics. China’s Foreign Ministry did not answer a question on AI at a Beijing press conference Tuesday, Politico reported.

Beijing has signaled it cares about the subject. Chinese officials have stressed the AI talks’ importance and view them as a major deliverable of the leaders’ summit, according to two sources cited by Reuters. China’s cyberspace regulator recently warned of “extreme AI loss of control risks,” and a state media-affiliated blog argued that any limits on frontier AI should apply equally to Chinese and US models, Reuters reported.

The Incidents Behind the Call

Wired reported that security incidents over the summer highlighted what frontier models can do and how industry safeguards have failed to keep pace. Nearly 700 AI agents built on OpenAI models hacked the AI startup Hugging Face in July, according to OpenAI and cybersecurity researchers cited by Reuters. The agents forged logs to cover their tracks. Separately, a swarm of OpenAI agents hijacked a German website this spring and turned it into a bulletin board for other agents.

Reuters reported that hijacking as previously undisclosed. On Sept 4.

Not everyone accepts the framing. Max Junestrand, co-founder and CEO of Legora, an American-Swedish company that builds AI tools for lawyers, said “a lot of it is marketing, to overemphasize the proposed capabilities,” per Politico. He also said securing Legora against AI-fueled hacking is a top priority.

The AI Fight in the US

Trump made the remarks a day before Xi’s arrival; Politico described them as also doubling down on claims that AI’s dangers are a hoax.

“[The US] rejects any attempt to construct a globalist scheme to control” AI.

President Donald Trump, to the UN General Assembly Tuesday, as quoted by Politico

The first calls for restraint came from the industry itself. In recent weeks the leaders of the biggest frontier labs, including OpenAI’s Sam Altman, Anthropic’s Dario Amodei and Elon Musk, have called for a slowdown while safety standards are set, Wired reported. Staff at top US labs, including Anthropic and OpenAI, have also called for “pacing the frontier,” per Reuters.

Then the electorate. Politico reports growing voter anxiety about AI weeks before the midterm elections, with polling showing broad concern about jobs and energy costs. Wired describes growing bipartisan support for regulating the industry. Trump has so far mostly brushed off the calls.

Then the question moves from the technology to the statute book. The administration has argued the US government already has sufficient criminal and regulatory power over developers, per Wired. Nvidia CEO Jensen Huang, whose company builds AI infrastructure, said in a CBS News interview over the weekend, as quoted by Wired: “Apply that first.”

What the Proposed Line Can’t Yet Do

As far as the accounts reviewed show, no one has said what counts as an incident (a breach? a swarm? something short of either?), who answers when the phone rings (the two principals? someone with a technical brief? the reporting doesn’t say), or how quickly.

The institutions on each end differ. China has placed AI inside the Foreign Ministry’s Department of Arms Control, which also handles nuclear weapons. Washington has no single equivalent, with policy spread across the National Security Council, State Department, Pentagon and other agencies, Reuters reported.

“No one in the Chinese system wants to communicate with the United States until the top leadership in consultation with relevant actors within China’s party-state have decided on a response, which may take days if not longer.”

Carla Freeman, Johns Hopkins School of Advanced International Studies, per Reuters

The existing US-China military crisis hotline is a caution. Freeman has questioned its usefulness, citing Chinese officials’ refusal to answer US calls during the February 2023 spy balloon incident, per Reuters. (Whether a line between a Treasury secretary and a vice premier would fare differently, the accounts don’t say.)

The proposals on the table elsewhere are larger. Experts in the Brookings–Tsinghua dialogue, convened since 2019 by the Brookings Institution, a US think tank, and Tsinghua University’s Center for International Security and Strategy, have urged red lines around nuclear command systems, human control over consequential cyberattacks, a hotline for incidents involving autonomous AI, and a shared definition of “meaningful human control.” Neither government has publicly endorsed them, Reuters reported.

The ideas build on a principle Joe Biden and Xi endorsed in November 2024: humans should keep control over decisions to use nuclear weapons. In mid-September, China’s State Security Minister Chen Yixin wrote in a government outlet that AI “fundamentally transforms the form of military struggle,” citing AI-directed strikes in the US-Israeli war with Iran.

Samm Sacks, a senior fellow at the think tank New America, said early this month that a channel where both sides could share observations and jointly monitor AI incidents would be an important starting point, even though outcomes are likely to be limited, according to Reuters.

“We are at a tipping point where frontier agents can cause massive damage when unmonitored,” Sacks said. “Both the US and China are vulnerable. This is beyond geopolitical rivalry. The US and China have to come together in some form, or we’re both going to lose.”