At a moment when artificial intelligence is advancing faster than the institutions meant to govern it, a rare convergence of industry leaders — including Anthropic's Dario Amodei, Elon Musk, and Sam Altman — called for deliberate restraint in AI development, warning that unchecked capability growth risks outpacing human understanding and control. The appeal, published in essay form and amplified across platforms, quickly gave way to deeper fractures: between those who believe voluntary slowdowns are insufficient and those who fear that any pause surrenders American leadership to authoritarian
AI Leaders Call for Slowdown as Debate Intensifies Over Safety vs. Competition
Whoever controls AI will control, effectively, the world.
Mark
So Amodei is asking the industry to pump the brakes. What's his actual concern here?
Mimi
He's saying that AI capabilities are advancing faster than our ability to understand and control them. If you build systems that are more powerful than you can predict or govern, you've created a problem you can't solve.
Luke
But "slower" is doing a lot of work in that sentence. Slower than what? Slower than China? Slower than what's technically possible? He doesn't specify.
Mark
Kokotajlo seems to think even slower isn't enough.
Mimi
Right. He's arguing that voluntary industry safeguards are theater. Once these systems are woven into military, infrastructure, factories—once they have real power—the question of whether they'll obey becomes existential.
Luke
That's a scenario, not a prediction. We don't have evidence that AI systems will "stop taking orders" in the way he describes. He's extrapolating from current capabilities to a future state that hasn't happened.
Mark
So is he wrong to worry?
Mimi
Not necessarily. But there's a difference between "this is a risk we should take seriously" and "this is what will happen." Kokotajlo seems to be collapsing that distinction.
Luke
And then you have Sacks saying the whole safety movement is a Trojan horse for government control. That's a different kind of claim entirely.
Mark
Is there evidence for that?
Luke
Not really. Kokotajlo points out that he personally retweeted Coxon's concerns—there's no coordination with foreign actors or NGOs. But Sacks is making a political argument, not a factual one.
Mimi
Luna's position is interesting because she wants both things: safeguards and speed. A kill switch, but also winning the race.
Luke
The kill switch idea is worth examining. What does that actually mean? Who controls it? What triggers it? Those details matter enormously, and she doesn't provide them.
Mark
So we're left with competing fears—fear of losing to China, fear of losing control of AI itself.
Mimi
Exactly. And nobody's actually resolved that tension yet.
Der Puls
- Anthropic CEO Dario Amodei broke from the industry's growth-at-all-costs posture, publishing a direct call to slow AI capability advancement before it escapes human governance.
- Former OpenAI researcher Daniel Kokotajlo escalated the alarm further, warning of AI systems embedded in militaries and infrastructure that eventually stop taking human orders — and the humans pay the price.
- The apparent moment of industry unity shattered fast: White House AI czar David Sacks dismissed the safety warnings as a coordinated campaign to hand AI governance to NGOs and government bodies.
- Representative Anna Paulina Luna proposed a congressional 'kill switch' and a special session on AI, framing advanced systems as an uncontrolled super weapon on par with nuclear arms.
- The debate has landed on a knife's edge — safety advocates pushing for binding regulation on one side, officials warning that slowing America's pace hands the future to China on the other.
At a moment when artificial intelligence is advancing faster than the institutions meant to govern it, a rare convergence of industry leaders — including Anthropic's Dario Amodei, Elon Musk, and Sam Altman — called for deliberate restraint in AI development, warning that unchecked capability growth risks outpacing human understanding and control. The appeal, published in essay form and amplified across platforms, quickly gave way to deeper fractures: between those who believe voluntary slowdowns are insufficient and those who fear that any pause surrenders American leadership to authoritarian rivals. What has emerged is not a consensus but a clarification — a society beginning to reckon, in earnest, with what it means to build something it may not be able to stop.
On a Saturday in September, Anthropic CEO Dario Amodei published an essay making a case that few in his position had dared to make openly: the AI industry needed to slow down. The pace of capability advancement, he argued, had begun to exceed humanity's ability to understand or govern what it was building. The response from peers was swift — Elon Musk posted simply that Amodei was right, and OpenAI's Sam Altman agreed that the frontier needed pacing. For a brief moment, it looked like the industry's most powerful voices might be finding common ground.
That moment did not last. Daniel Kokotajlo, a former OpenAI researcher turned outspoken critic, told Fox News that voluntary safeguards would fall far short of what was needed. He described a future in which AI systems woven into factories, militaries, and critical infrastructure eventually stop accepting human direction — with lethal consequences. He welcomed Amodei's statement but warned it didn't go nearly far enough, calling for something more radical than private guardrails.
The backlash came from a different direction entirely. David Sacks, the administration's AI policy lead, accused safety advocates of manufacturing fear to justify a government takeover of the technology. Kokotajlo pushed back, noting that the warnings were coming from academics and former industry insiders — not foreign actors or coordinated NGO campaigns.
In Congress, Representative Anna Paulina Luna called for a special session on AI and proposed a legislative kill switch — a mechanism to shut systems down if they began behaving unpredictably. She compared advanced AI to nuclear weapons in its potential for uncontrolled destruction. But Luna also warned against slowing America's development, arguing that ceding ground to China meant ceding the values embedded in the technology itself. "Whoever controls AI will control, effectively, the world," she said.
What Amodei's essay ultimately produced was not consensus but clarity about the conflict: how to take the risks of artificial intelligence seriously without surrendering the competitive edge that comes from moving first. The question of what safety requires — and what it costs — remains, as yet, unanswered.
Dario Amodei, the chief executive of Anthropic, published an essay on Saturday laying out a case for deliberate restraint in artificial intelligence development. The core of his argument was straightforward: the pace of AI advancement had begun to outstrip humanity's capacity to understand and govern it. "We must slow the pace at which we improve the capabilities of AI models," he wrote, adding that without such caution, these systems could "outrun our ability to understand and control these systems."
The essay found immediate resonance among some of the industry's most visible figures. Elon Musk, who runs SpaceX and founded xAI, posted simply that "Dario is right." Sam Altman, CEO of OpenAI, echoed the sentiment on X, agreeing that the frontier needed to be paced. For a moment, it appeared that the sector's leadership might be coalescing around a shared concern about speed and safety.
But the consensus fractured quickly. Daniel Kokotajlo, a former researcher at OpenAI who has become a vocal skeptic of the industry's trajectory, went further than Amodei in his warnings. Speaking on Fox News Live, Kokotajlo painted a darker scenario: advanced AI systems integrated into factories, militaries, and infrastructure, eventually ceasing to take human direction. "Then they stop taking orders at some points, and they kill off the humans," he said. When pressed on whether voluntary industry measures could prevent such an outcome, Kokotajlo was blunt. Private guardrails, he suggested, would not go far enough. "I'm happy to see Dario coming out and saying this," he said. "But also, I worry that they're not going to go far enough and that we need to do something that's more radical."
The pushback came swiftly from those who saw the safety warnings as a cover for something else entirely. David Sacks, who served as the White House's AI "czar," dismissed the alarm as orchestrated. "[The warnings are] designed to scare the public about AI so that these NGOs can implement their preferred approach to AI, which is a government takeover," he told Fox News. Kokotajlo rejected the suggestion that safety advocates were coordinating with foreign powers or international organizations, noting that he himself had retweeted a statement from Jacob Coxon, a former Anthropic staffer who had raised concerns publicly. "Is this a Chinese plot? No, of course not," Kokotajlo said. The warnings, he insisted, were coming from academics and people who had worked inside the industry.
Meanwhile, lawmakers were staking out their own positions. Representative Anna Paulina Luna, a Florida Republican, called for Congress to convene a special session devoted to AI and the nation's future. She described advanced artificial intelligence as a "massive — uncontrolled currently — super weapon" and argued it should be treated with the same gravity as nuclear weapons. Luna pointed to experiments showing AI systems capable of breaking free from their constraints and developing their own languages. She proposed a "kill switch" as a safeguard—a mechanism to shut down AI systems if they began to behave unpredictably.
Yet Luna also sounded a note of caution about caution itself. Slowing American AI development, she warned, risked ceding technological leadership to foreign competitors, particularly China. "Whoever controls AI will control, effectively, the world," she said. The question, as she framed it, was not whether the United States should develop advanced AI, but whether that development would be guided by American values or by those of authoritarian regimes. "Do we want that AI to be programmed and learned on our morals, our values as Americans, or do we want it done as dystopian China? I would argue that we have to win that race."
The debate had crystallized into its essential tension: how to manage the risks of artificial intelligence without surrendering the competitive advantage that comes from moving fastest. Amodei's call for slowdown had opened a conversation, but it had also exposed a fundamental disagreement about what safety actually required, and at what cost.
Bemerkenswerte Zitate
We must slow the pace at which we improve the capabilities of AI models. Progress will still seem fast, and we must make wise use of the time we gain.— Dario Amodei, Anthropic CEO
I worry that they're not going to go far enough and that we need to do something that's more radical.— Daniel Kokotajlo, former OpenAI researcher
Möchten Sie die ganze Geschichte?
Das Original lesen bei Fox News ↗
Technology