AI bosses are calling for a slowdown. Nobody knows what that means.

Key points

• Anthropic CEO Dario Amodei publicly called for a slower pace of AI development on Saturday; OpenAI's Sam Altman and Elon Musk both said they agreed.
• Jacob Coxon, a researcher who recently quit Anthropic, told BBC Technology on Sunday that staff building these systems are "genuinely frightened" for humanity's future.
• President Trump declared on Sunday that "whoever wins AI, wins," signalling the US government has no appetite for pausing its race against China.
• Amodei's three-point plan proposes independent monitoring of AI models, industry-wide rules and global regulation, but no part of it has been spelled out in workable detail.
• No international body currently has the authority or the tools to enforce an AI slowdown across competing nations.

The people building the most powerful AI systems in the world are now saying, out loud, that they are scared of what comes next. That is not a small thing.

On Saturday, Dario Amodei, the chief executive of Anthropic, published a call to slow the pace of AI development. By Sunday, two rivals had agreed: Sam Altman of OpenAI and Elon Musk of xAI. A researcher who recently left Anthropic told the BBC that colleagues still inside the company are "genuinely frightened" for the future of humanity. We first covered this chorus of concern on 14 September, when the commitments were already drawing scepticism.

So why can't the industry just hit pause?

Because slowing down is easy to say and almost impossible to do. The core problem is geopolitical: America believes it is in a race with China to build the most powerful AI, and no American company wants to be the one that blinks first.

President Trump put it bluntly on Sunday: "Whoever wins AI, wins." Every industry insider the BBC spoke to made the same point: even if their company paused, rivals would not. The comparison to Cold War nuclear disarmament is hard to avoid. Nobody wanted to go first then either.

Amodei proposed a framework covering independent expert oversight of models as they are built, sector-wide rules, and some form of global agreement. Critics were not impressed. "Nobody has given a substantive explanation of what 'slowdown' means," said Ed Zitron, chief executive of EZ Primary Research. Stopping new AI training, he warned, would freeze American products while Chinese labs caught up.

What does this mean for ordinary people?

Governments are actively encouraging AI adoption: in the UK, the technology features in NHS rollout plans and sits at the centre of the government's economic growth strategy. A former adviser told the BBC plainly: "There is no plan B."

The industry is also burning through investor money faster than it earns revenue, and economists widely expect not every current giant to survive. Those that do could become the most powerful corporations the world has ever seen, which brings its own problems.

Professor Dame Wendy Hall, a computer scientist who has advised the United Nations on AI, frames the moment with a farmer's analogy: let a bull escape, and you cannot blame the bull for the damage. "Of course it's not the bull's fault," she told the BBC. The fences were not strong enough, and that is exactly what weak AI safety guardrails look like.

Sasha Luccioni, founder of Sustainable AI, puts her worry elsewhere: "I'm not worried about the existential risks of AI, I'm worried about the corporate greed of the companies that are creating it."

That gap, between genuine fear at the top and the financial pressure to keep accelerating, is what makes this moment so hard to read. The calls for a slowdown are real. The mechanism to achieve one does not yet exist, and the window to build one is narrowing fast.