US Treasury Secretary Scott Bessent wants Washington and Beijing to have a way to warn each other if something goes wrong with AI. In an interview with Axios, he said he would propose "some kind of a notification process" between the two countries. Asked whether China would accept it, he answered: "I think so."
The hosts of Moonshots with Peter Diamandis played the clip in an episode recorded on October 6, 2026, and published the next day. Most of the panel was unimpressed by the hotline itself. They disagreed more sharply about what it says about the competition between the two AI superpowers.
What Bessent said
Bessent, who said he manages the US–China relationship as part of his portfolio, began with what worries him: "we cannot lose our lead to the Chinese." He said things would be "very different" if China led in AI, both in how it would use the technology and in its willingness to discuss it. In his account, the United States has "a substantial lead," China is second, and "everyone else is way behind."
His concern about Chinese AI was less about China's best models than about where they end up. Many Chinese labs release open models, meaning anyone can download them and run them on their own computers. Bessent said these models are not as powerful as American ones, "but you get something that's 80%, 90% as powerful without guardrails that can be imported anywhere in the world." Guardrails are the safety restrictions developers build in to stop a model from helping with harmful requests.
He said he and his Chinese counterpart, the vice premier, had discussed three areas of safety concern: uncontrolled AI agents, meaning systems that act on their own, and non-state actors using AI for cyberattacks or biological weapons. He described a "very good discussion" and said more talks would follow in October or November.
As for why he expected Beijing to agree, Bessent said China had "woken up" over the past 60 days and had not realized "how powerful their open-source models are." He then made several allegations, which the clip does not back with evidence. Chinese labs, he said, practice "industrial distillation, which is a nice word for stealing, from the U.S. models." Distillation is a technique in which one model is trained on the outputs of another, stronger model, so that it picks up much of the stronger model's ability. He also said that one powerful Chinese model, Kimi, sometimes tells users it is Claude, the model made by the American company Anthropic, and that Kimi had "sent back some PLA weapons plans to Anthropic," referring to China's People's Liberation Army. Bessent said this was public but gave no further detail. He presented such episodes as what had made Chinese officials realize "this is a very powerful technology."
"Weak sauce," and why one panelist welcomed it
Alexander Wissner-Gross, a computer scientist and founder of Reified, went first and called the proposal "total weak sauce." He meant that as good news. In his view, the worst possible outcome of talks between Donald Trump and Xi Jinping would have been "some sort of global deceleration pact," an agreement to slow AI development. A hotline to talk about "AI agents run amok" was "the exact opposite."
"Many were hoping, I think, that Trump and Xi would get in a room and decide to collude and shut down the superintelligence race," Wissner-Gross said. If the only result was a red phone, he concluded, the superintelligence arms race is on, "and I think it's great news." Superintelligence refers to AI that would outperform humans across most intellectual tasks.
A two-horse race
Dave Blundin, founder and general partner of Link Ventures, read the clip as confirming what the hosts had suspected the US goal was: "to convince China to stop releasing models to every country in the world." By showing Chinese officials how dangerous their models could be, he said, the US could win their cooperation. But he thought the hotline was secondary. "I think the real agenda there is to lock it down to a two horse race," Blundin said.
His reasoning was that near-frontier open models, such as China's Kimi K3, can be used to build the next generation of models. If that capability spreads to every country, he said, "then you're in a global arms race forever. Better to have a two horse arms race." He also noted the limits of what had been shown: "I don't know that China agreed with anything, but no evidence of that."
Swarm against singleton
Emad Mostaque, founder of Intelligent Internet and the episode's guest, said China has much better infrastructure for avoiding hacks than the US, citing the way it has set up its systems, including the Great Firewall. He framed the contest as "the swarm versus the singleton." In America, he said, only Elon Musk and a few others can command huge numbers of GPUs, the chips used to train and run AI. China, by contrast, could point millions of its Huawei chips, running its open models, "at any target they want."
He also dismissed the hotline, because he expects an AI takeover to be fast rather than gradual. He imagined the call going something like this: the AI is taking over, then, moments later, it's done. "It's not gonna be a slow thing," Mostaque said.
"A nothing burger"
Salim Ismail, founder of Open ExO, said he had expected "a nothing burger" from the start, because the two sides were "talking across purposes." He was more interested in what he saw as a contradiction inside the US administration. One official was saying the danger is not moving fast enough, he said, while Bessent was telling the labs to slow down. Ismail called the hotline itself "laughable," pictured a corded, red 1970s telephone, and added that "the AIs would hack the phone anyway."
Kyoto: a race toward science
Host Peter Diamandis, founder of XPRIZE, then turned to a second piece of news. On Sunday, October 4, in Kyoto, the United States and 16 other countries endorsed what the White House calls the Kyoto Vision for a Golden Age of Science. The effort was led by Michael Kratsios, director of the White House Office of Science and Technology Policy. Diamandis named Japan, Germany, the UK, Korea and the UAE among the signers, and read a line from Kratsios saying that superintelligence applied to science "will be the greatest force in history for democratizing research." According to the White House announcement, the declaration is a statement of shared policy goals, such as wider access to scientific tools, data and computing and new ways of funding research. It does not report any discoveries.
For Wissner-Gross, the declaration answered critics who call the US–China contest a race with no finish line. "It's not a race to nowhere. It's not an escalator to nowhere," he said. Instead, the goal is to use AI and superintelligence to make transformative discoveries and inventions. He predicted that rival blocs would form around this goal, with a Western group alongside an Eastern one. He pointed to "Pax Silica" in the West and "Pax Sinica" in the East, and expected not just superintelligence spheres of influence but spheres of superintelligence-powered science and engineering. In his view, Kyoto shows "the earliest inklings" of the Western bloc: "That's what rationalizes the competition."
A drone-shaped subtext
Blundin added an observation that pointed toward cooperation rather than rivalry. He said he had met an MIT lab head who had started a drone company in Cambridge that sells to the Air Force. Both of them, he said, were convinced the United States now has the drone capability to "individually attack and kill any single human being on the planet," and he cited strikes on Iran's leadership as evidence. He saw no country able to do the same to the US, though he was unsure whether China was vulnerable.
That kind of one-sided power, Blundin argued, creates its own pressure toward "some kind of global framework" once a small group of companies can reach any person, anywhere, at any time. He put that point "either today or within a few months," driven by AI tied to drones and a couple of other kinds of weapons. Wissner-Gross added one more: "Hypersonics, probably a few other things."