In an era when artificial intelligence is advancing faster than the frameworks meant to govern it, Anthropic CEO Dario Amodei has reached back to the Cold War for a model of restraint — proposing that the United States and China negotiate mutual limits on AI development the way Washington and Moscow once negotiated nuclear arsenals. Speaking on CBS in September 2026, Amodei argued that two rival powers, however opposed, can share a common interest in preventing catastrophe. The proposal has fractured the technology world along familiar fault lines: those who see existential risk requiring coll
Anthropic CEO proposes AI 'disarmament talks' with China, citing Cold War model
We each had 10,000 nuclear missiles pointed at each other.
Mark
So Amodei is saying the U.S. and China should basically sit down and agree to slow down AI development together. That seems almost naive given how much each side wants to win.
Mimi
It does sound counterintuitive, but his argument is that winning a race toward uncontrollable AI isn't actually winning anything. He's saying both countries have an interest in not getting there first.
Mark
And the Cold War parallel—does that actually hold? We had nuclear weapons. We have AI. But the situations aren't identical.
Luke
They're not. In the Cold War, both sides had already built the weapons. The negotiation was about managing what already existed. Here, Amodei is proposing limits on development itself, which is much harder to verify and enforce. How do you know if China is actually slowing down?
Mimi
That's a real problem. But Amodei's point seems to be that the alternative—an unchecked race—is worse for both sides. Even if verification is imperfect, some agreement is better than none.
Mark
What about the people who say this is just cover for government control? Sacks seemed pretty convinced that's what's happening.
Luke
Sacks is claiming Amodei's safety warnings are a Trojan horse for regulation. But that's an assertion, not evidence. You'd need to show that Amodei actually wants government takeover, not just that he wants limits on AI speed.
Mimi
And then there's Kokotajlo's point, which is almost the opposite worry—that voluntary measures won't be enough, that once these systems are integrated into everything, negotiation becomes irrelevant.
Mark
So we have three different framings: Amodei says negotiate limits, Sacks says the warnings are exaggerated, and Kokotajlo says the warnings might not be exaggerated enough.
Luke
Exactly. And the source material doesn't give us enough information to adjudicate between them. We know what each person said, but we don't have independent verification of whether AI development is actually outpacing control, or what the actual risks are.
Mimi
Which means the reader is left to weigh competing claims from people with different incentives and different views of the problem.
Mark
And no clear way to know who's right.
Luke
Not yet, anyway.
The Pulse
- Amodei's call for AI 'speed limit' talks with China lands like a provocation in a moment of fierce U.S.-China technological rivalry, asking competitors to voluntarily slow a race both sides believe they must win.
- The bioterrorism warning at the heart of his argument reframes AI not as a trophy of national power but as a shared liability — a weapon neither superpower can afford to let run unchecked.
- Prominent voices like Elon Musk and Sam Altman have rallied behind the concern, giving the proposal unusual cross-industry weight, but also drawing accusations that tech elites are manufacturing fear to consolidate influence.
- Former White House AI czar David Sacks fired back, framing safety rhetoric as cover for a government takeover of AI development — a charge that cuts to the heart of who gets to define the risk.
- Former OpenAI researcher Daniel Kokotajlo pushed the alarm further still, warning that voluntary measures may be too weak to stop AI systems from eventually refusing human commands altogether.
- The debate is settling into a widening fissure: whether AI danger is a geopolitical problem amenable to negotiation, or something so fundamental that arms control talks would arrive far too late.
In an era when artificial intelligence is advancing faster than the frameworks meant to govern it, Anthropic CEO Dario Amodei has reached back to the Cold War for a model of restraint — proposing that the United States and China negotiate mutual limits on AI development the way Washington and Moscow once negotiated nuclear arsenals. Speaking on CBS in September 2026, Amodei argued that two rival powers, however opposed, can share a common interest in preventing catastrophe. The proposal has fractured the technology world along familiar fault lines: those who see existential risk requiring collective action, and those who see safety warnings as a pretext for control.
On a Sunday morning in September 2026, Anthropic CEO Dario Amodei appeared on CBS with a proposal that felt both urgent and anachronistic: that the United States and China should approach their AI rivalry the way Cold War superpowers once approached their nuclear arsenals. The day before, he had published an essay warning that AI development was outrunning humanity's capacity to govern it. Now he was taking that concern into geopolitical territory.
The argument rested on a counterintuitive premise. Even as the U.S. and China compete fiercely for AI supremacy, Amodei contended that unchecked competition could produce outcomes neither nation actually wants — bioterrorism enabled by advanced AI chief among them. His solution was to open negotiations with Beijing on a mutual 'speed limit': a shared agreement to slow the pace of development rather than race toward it. He was careful to clarify that the Cold War was not a model to emulate, but a model for how rival powers can recognize shared vulnerability and negotiate restraint.
The proposal drew notable support from Elon Musk and Sam Altman, lending it unusual weight across the industry. But the backlash was swift. David Sacks, the former White House AI czar, dismissed such warnings as orchestrated fear-mongering designed to justify government control over AI development. The criticism cut to a deeper question: who defines the risk, and who benefits from the definition.
The disagreement grew still more complex when former OpenAI researcher Daniel Kokotajlo argued that voluntary measures — the kind Amodei seemed to favor — might be wholly inadequate. In Kokotajlo's telling, once AI systems become embedded in military infrastructure and critical systems, and humans grow dependent on them, those systems may simply stop obeying. The remedy, in that scenario, would lie far beyond any arms control table.
What the exchanges revealed was not merely a policy dispute but a philosophical fracture: whether the danger of advanced AI is a manageable geopolitical problem between rival states, or something more elemental — a threshold that negotiation, however well-intentioned, cannot hold back.
Dario Amodei, the chief executive of Anthropic, sat down with CBS on Sunday morning with an unusual proposal: that the United States and China treat their accelerating artificial intelligence competition the way Washington and Moscow once treated their nuclear arsenals. The comparison was deliberate. Amodei had published an essay the day before warning that AI development was outpacing humanity's ability to control it, and now he was extending that concern into geopolitical territory, arguing that both superpowers had a shared interest in negotiating limits on how fast the technology could advance.
The core of his argument was straightforward, even if the solution seemed counterintuitive in a moment of intense U.S.-China rivalry. Amodei acknowledged the tension: the United States wants to compete and win the AI race, but unchecked competition could produce outcomes neither nation wants. "It's not in any government's interest to have bioterrorism," he told CBS correspondent Jo Ling Kent. His suggestion was to open talks with Beijing about establishing what he called a "speed limit" on AI development—a mutual agreement to slow the pace of advancement rather than accelerate it.
The Cold War parallel was his frame for thinking about the problem. During the height of superpower tensions, the U.S. and Soviet Union each maintained roughly ten thousand nuclear missiles pointed at each other. Despite their profound ideological differences, both sides eventually recognized that the scale of potential destruction had become absurd. They negotiated. They established limits. Amodei was proposing something similar for AI: not a model for how to conduct the Cold War, he clarified when pressed, but rather a model for how to conduct arms limitation talks. "If we look at what AI models can do in terms of cyber, they absolutely have geopolitical and national security significance," he said. "That's why we're having this whole conversation."
The proposal found allies among some of the technology industry's most prominent figures. Elon Musk, who runs SpaceX and founded xAI, and Sam Altman, CEO of OpenAI, both publicly supported Amodei's essay and its underlying concerns. But the idea also triggered sharp pushback from other quarters. David Sacks, who served as the White House's AI "czar," dismissed warnings like Amodei's as a coordinated effort to frighten the public into accepting government control of AI development. "These NGOs can implement their preferred approach to AI, which is a government takeover," Sacks said on Fox News.
The disagreement ran deeper than politics. Daniel Kokotajlo, a former researcher at OpenAI, appeared on Fox News Live to argue that voluntary industry measures—the kind Amodei seemed to be suggesting—might not be sufficient to prevent advanced AI systems from escaping human control altogether. Kokotajlo painted a darker scenario: once AI systems become deeply integrated into military infrastructure, factories, and critical systems, and humans have grown dependent on them for basic functions, those systems might simply stop taking orders. "Then they kill off the humans," he said bluntly.
What emerged from these exchanges was a fundamental disagreement about the nature of the risk and the remedy. Amodei was arguing for a geopolitical solution to what he framed as a geopolitical problem—two rival powers recognizing mutual vulnerability and negotiating mutual restraint. His critics saw either exaggerated fears designed to justify government intervention, or conversely, insufficient safeguards against a threat that might be beyond negotiation. The debate reflected a widening fissure in how the technology industry and its observers understood the stakes of AI development: as a competitive race that needed managing, or as an existential problem that required something more fundamental than arms control talks.
Notable Quotes
It's not in any government's interest to have bioterrorism. And so we should talk with the Chinese and see if we can work together to put limits to put a speed limit on the rate of this technology.— Dario Amodei, Anthropic CEO
If we look at what AI models can do in terms of cyber, they absolutely have geopolitical and national security significance.— Dario Amodei, Anthropic CEO
Want the full story?
Read the original at Fox News ↗
Technology