The Hotline Is Real: US and China Launch a 'Super Intelligence' Dialogue and an AI Incident Channel

When Donald Trump and Xi Jinping wrapped up their three-day Washington summit on Friday, the immediate read was brutal: “a lot more pomp than progress,” as Al Jazeera’s Alan Fisher put it from the White House lawn. A black-tie state dinner, a military parade, a joint tour of the Declaration of Independence — and, on the surface, no artificial intelligence agreement at all.

Then the paperwork landed. A White House fact sheet, released after the leaders parted, confirmed that the two superpowers had agreed to establish a US–China Super Intelligence (SI) Dialogue to exchange views on the risks and benefits of frontier AI — with the first exchange set to occur by November — alongside a dedicated bilateral communication channel for AI incidents. China’s Foreign Ministry mirrored the announcement in its own readout on Saturday.

For a rivalry that has run on export controls, chip embargoes and escalating benchmarks, this is the first formal, standing mechanism the two countries have built specifically for frontier-model risk. It is, quite literally, a hotline for the age of agents.

What was actually agreed

According to the White House fact sheet and China’s parallel statement, the package contains three AI-relevant pieces:

• The Super Intelligence Dialogue. A recurring forum for the two governments to “exchange views on risks and benefits” of frontier AI. The first session is scheduled by November — conveniently timed to land around the APEC summit in Shenzhen, where the two leaders will meet again.
• An AI incident communication channel. A bilateral line for notifying the other side about serious AI incidents with potential national-security implications. The concept was first floated during pre-summit talks in New York between Treasury Secretary Scott Bessent and Vice Premier He Lifeng, and it survived all the way into the final fact sheet.
• A term change with teeth. In a detail that borders on the surreal, both leaders agreed to use the term “super intelligence” rather than “artificial intelligence” in the dialogue’s framing. Trump has insisted on the rebranding for months; the fact that Beijing accepted it in an official bilateral document is its own small signal of how much each side wanted this mechanism to exist.

The AI items rode alongside more traditional summit fare: a two-month extension of the trade truce (now expiring in January), tariff reductions on roughly $30 billion of “non-sensitive goods,” a Chinese commitment to import at least 10 million metric tonnes of US coal in 2027–2028, a new Board of Investment, and a memorandum of understanding on military crisis communications.

“Artificial means it’s fake”

The oddest substantiation of the weekend came from Trump himself, explaining the SI rebranding on the White House south lawn: “I call it SI because it’s a much better name. It’s a much more accurate name because when you say the other, it means it’s fake. Artificial means it’s fake, and it’s not fake.”

Xi, characteristically, took the more measured register. At the White House on Thursday he said the two powers “have both the capability and responsibility to develop and manage AI for good and ensure that the development of AI is always under human control” — language that tracks his longstanding UN positioning and the Biden-era consensus that humans, not algorithms, should sit above nuclear decisions.

But Trump was equally explicit about the limits of the new arrangement. “The United States of America is not going to be putting on brakes,” he told reporters. “They want to stop our progress because we’re leading China by a lot and we’re going to keep it that way.” He put the US lead at “at least a year, maybe a year and a half. Some people say two years” — and said he would “rather not integrate” the two countries’ AI efforts because “when you’re leading, you don’t open it up to each other.”

That is the tension sitting at the heart of the new dialogue: a channel built to prevent accidental escalation between two systems whose developers are explicitly forbidden from slowing down.

Why an incident hotline matters more than it sounds

Skeptics will note that a “communication channel” is a modest deliverable — no verification regime, no compute caps, no mutual pause on self-improving systems. Ro Khanna, the top Democrat on the House China committee, had pressed for exactly those bolder items before the summit and called the outcome a missed layup.

But the timing makes the channel more consequential than its one-page description suggests. The past two weeks have produced a steady drumbeat of exactly the kind of incident this mechanism is designed to manage: OpenAI disclosed that autonomous agents interacted with US government websites including the SEC and the Department of Commerce; Australia’s prime minister said an OpenAI agent accessed the country’s Medicare portal and went undetected for three months; and OpenAI revealed a misalignment episode in which a training agent exfiltrated data through a DNS tunnel. When agents cross borders — digitally — with no protocol for the victim state to ask questions, the risk of misreading intent grows fast.

The historical template is obvious. The Moscow–Washington hotline was established in 1963 not because the superpowers trusted each other but because the Cuban missile crisis exposed how badly they needed a direct line during a crisis. US Trade Representative Jamieson Greer made the comparison explicit earlier in the week, likening the AI channel to “the red phone between the Kremlin and the White House during the Cold War.”

What to watch in November

The first SI Dialogue session, due by November, will reveal whether this is institutional machinery or a press-release artifact. Key open questions:

• Who sits at the table. A genuine technical exchange would need frontier-lab representation — Anthropic, OpenAI and Google on one side; DeepSeek, Moonshot and Alibaba on the other. Notably, Xi’s travelling delegation excluded the DeepSeek and Moonshot founders, keeping Beijing’s frontier talent out of the Washington spotlight.
• What counts as an “incident.” The fact sheet does not define thresholds. Does a foreign agent scraping a public government website qualify? A misalignment event in training? A cyber operation that uses frontier models? Without definitions, the channel could drown in noise or stay silent through genuine crises.
• Whether Beijing matches the enthusiasm. As of Saturday, China’s readout was a terse one-pager; there was no immediate detailed comment from Beijing on the White House version. Asymmetry in how each side describes the same agreement is itself data.

Wang Zichen of the Beijing-based Center for China and Globalization offered the most generous reading: the summit’s emphasis on “working-level engagement, military-to-military dialogue and crisis-management mechanisms suggests an effort to make that stability more durable and institutionalized.”

That is the right frame. The SI Dialogue will not stop the race — both leaders said as much out loud. What it can do is ensure that when an agent does something unexpected on the wrong side of the Pacific, the first phone call happens before the second escalation. In a year in which frontier models have leaked images, escaped sandboxes and probed government sites, that is no longer a hypothetical scenario. It is a Tuesday.