The people who run the biggest AI labs now say they should hit the brakes. Anthropic CEO Dario Amodei posted an essay calling for a global slowdown in how fast frontier AI models get more powerful, and within hours OpenAI CEO Sam Altman, xAI's Elon Musk, and Google DeepMind chief Demis Hassabis backed the idea online. That kind of unity from rivals is rare. It landed right after a former Anthropic researcher, Jacob Coxon, resigned and warned that some inside these companies genuinely believe AI could kill all humans by the end of the decade.
The warning that lit the fuse
Coxon's post spread fast because it did not sound like a press release. He wrote that the people building AI earnestly believe it could kill us all by the end of the decade, and that this is not a marketing stunt. Minutes later, Evan Hubinger, Anthropic's head of alignment science, publicly agreed and said he personally sees more than a 10 percent chance of that happening within the next decade. Those lines moved the debate from abstract risk to something that felt immediate.
The claim is not that AI will definitely end humanity by 2030. It is that some of the people closest to the most advanced systems think the odds are high enough to treat the risk seriously, and that current safety work has not yet solved how to control superintelligence. Coxon added that highly capable systems could hack critical infrastructure, gain resources in the physical world, and speed up dangerous biological research if guardrails fail. Other experts push back hard. Gary Marcus has argued that the idea of AI eliminating all humans in five years is preposterous, and that the chance is near zero. The 2026 International AI Safety Report says current systems are not yet capable of overtaking humans, even as capabilities improve quickly.
What the CEOs are proposing
Amodei's plan is simple to state and hard to pull off. He wants frontier AI labs to slow the pace at which they improve model capabilities, embed independent safety evaluators with deep access inside their companies, agree on common safety standards across democratic nations, and work toward international coordination that includes authoritarian governments like China. Altman said OpenAI welcomes a federal framework that sets consistent safety requirements for frontier AI and that no amount of American competitive pressure should justify recklessness. He also said he would match Amodei's commitment to outside evaluators.
Markets reacted right away. AI-linked stocks fell worldwide after the CEOs' comments, with reports noting the drop shook confidence in an industry whose infrastructure spending has helped drive record highs. OpenAI also said it will not proceed with an IPO this year, according to reporting on the same wave of news. At a San Francisco conference this week, Amodei reiterated the slowdown call while Nvidia's CEO argued against deceleration, showing the split between safety-first voices and those pushing to keep acceleration going.
The room this story lives in
Online, the temperature split along familiar lines. One side treated the warnings as a long-overdue admission that the race has moved faster than safety can keep up. The other side called it doom-mongering that distracts from real, present harms and makes sensible progress harder. What stood out was not the volume but the source. When the people building the systems say they fear losing control, even skeptics had to engage the substance instead of waving it off as hype.
There is still no scientific consensus that AI will wipe out humanity by 2030, or even that it will become powerful enough to pose that kind of threat. Surveys of experts have produced a wide range of estimates, with some putting the chance of extinction-level outcomes from advanced AI in the single digits over longer time frames. But the possibility is now credible enough that three of the leading AI lab chiefs are asking for coordinated restraint while their companies and partners continue to commit hundreds of billions to data centers and compute.
The core tension is plain. Slowing down buys time to build guardrails and test systems before they cross new capability thresholds. It also risks ceding ground to competitors who do not slow, whether other firms or other countries. Amodei's answer is to make safety a shared rule of the road, with outside auditors and cross-border standards so no one gets punished for going slower. Whether that can hold in a market that rewards speed is the open question.
For now, the most striking fact is not the doom language. It is that the same executives who have raced to ship bigger models are now asking, in public, to pace the frontier. The next move will show whether that request is a real shift or a pause long enough to catch breath before the next sprint.
More on: News