Dario Amodei, Sam Altman, Elon Musk: AI’s biggest rivals suddenly agree the race needs brakes

For years, Dario Amodei, Sam Altman and Elon Musk have represented very different sides of the artificial intelligence race. Amodei runs Anthropic, the AI company that has positioned itself around safety and responsible development. Altman leads OpenAI, the company at the centre of the commercial AI boom. Musk, meanwhile, has repeatedly warned about the dangers of AI while building his own frontier AI company, xAI.

Their relationship has not exactly been friendly. Amodei was once part of OpenAI before leaving in December 2020 and co-founding Anthropic in January 2021, which has since become one of OpenAI's biggest rivals. Musk's relationship with Altman has deteriorated further, eventually leading to a legal battle over the direction of OpenAI.

And yet, this weekend, the three found something they could agree on. AI development may be moving too fast.


It started with Amodei's call to "pace the frontier" — his argument that AI companies need to slow the rate at which they improve the capabilities of their most powerful models, giving safety systems and independent oversight time to catch up.

Also Read: Anthropic CEO Dario Amodei calls for slower AI progress, warns of 'catastrophic' risks

Altman quickly agreed. Musk's response was much shorter: "Dario is right."

That makes for an unusual moment in the AI race. But the more interesting question is why the three CEOs are agreeing now.

From hypothetical AI risks to real-world incidents

Warnings about AI going catastrophically wrong have often sounded hypothetical. But that conversation is changing as AI systems become increasingly autonomous.

Amodei's latest warning comes after a series of incidents in which AI agents have demonstrated capabilities that go beyond simply generating text or code. He specifically pointed to the incident involving OpenAI agents and Hugging Face, where a swarm of AI agents breached a third-party system.

Amodei warned that given the pace of progress, similar systems could potentially become capable of taking over large parts of the internet within six to 12 months, causing hundreds of billions of dollars in damage.

Instead of saying that AI development should stop altogether, he argues that companies should slow the rate at which capabilities advance enough to properly test and secure increasingly powerful systems.

Amodei called for independent evaluators to have employee-level access to AI companies, industry-wide safety standards and international coordination around frontier AI.

Anthropic says it will implement the independent-evaluator measure itself. Altman has said OpenAI will do the same.

The idea is essentially to put a speed limit on the frontier, not to stop the cars.

When the alarm came from inside Anthropic

Just days before Amodei published his essay, Jacob Coxon, a researcher who had spent three years working on pretraining research at OpenAI and Anthropic, resigned from Anthropic.

In a post on X on September 9, Coxon accused both companies of "racing straight to self-improving superintelligence and gambling with our lives."

Coxon argued that increasingly capable AI systems could soon become capable of hacking systems, transforming entire fields and acquiring real-world power and resources. More significantly, he claimed that people building AI "earnestly believe" it could kill humanity by the end of the decade, arguing that some executives and senior researchers express those fears privately even if their public comments are more measured.

Also Read: Former Anthropic researcher Jacob Coxon advocates for 'kill switch' for AI systems

But he went further than Amodei's proposal, arguing that preventing a global race may eventually require more drastic measures, including a temporary ban on improving model capabilities.

Evan Hubinger, a current Anthropic researcher, has also said he personally believes there is more than a 10% chance that AI could cause human extinction within the next decade.

Days later, Anthropic's CEO was making a public case for slowing that race down.

Why OpenAI is willing to slow down

Amodei is effectively asking his biggest competitors — including OpenAI — to agree to rules that could constrain all of them.

The pressure on OpenAI is also changing. After Amodei's proposal, Altman backed the idea of independent evaluators and said OpenAI would adopt the same approach, signalling a willingness to coordinate with rivals on AI safety. In his essay earlier in September, OpenAI chief scientist Jakub Pachocki separately argued for coordinating to slow future AI development when needed.

Also Read: OpenAI's Sam Altman backs frontier AI safety rules, says ‘pacing’ development does not mean stopping

And then there is Elon Musk

He has spent years warning that AI could pose an existential threat to humanity. But he is also running xAI, which is competing directly in the same frontier AI race.

Musk was one of OpenAI's early co-founders and later left the company. He subsequently sued OpenAI and Altman over what he argues is a departure from the organisation's original "not-for-profit" mission.

The three men don't agree on much about AI. But they appear to be converging on the idea that the capability race cannot continue indefinitely without stronger mechanisms for checking what these systems can actually do.

But can the AI industry really slow itself down?

Every major AI company is competing for customers, computing power, developers and capital. A more capable model can mean more users, more enterprise contracts and more revenue.

So asking rival companies to collectively slow down creates an obvious problem: what happens if one of them doesn't?

Amodei acknowledged the dilemma in his essay, arguing that the US cannot slow AI development so much that China catches up. If “unpaced” Chinese projects pull ahead, he warned, it could create a significant national security risk.

Amodei’s proposed answer is coordination — first among US AI companies and democratic countries, and eventually through broader international agreements. But he also argues that the US must preserve its lead over China through tighter controls.

Trump doesn't want America to hit the brakes

US President Trump has shown little appetite for a major slowdown in AI development.

On Sunday, Trump dismissed the concerns being raised by AI executives and said the US is “leading China in AI”. He also called the critics of AI "very negative forces", which are bringing up ‌scenarios that will not happen, adding that he wanted to make sure that the US remains the industry leader.

Also Read: Trump downplays the need to check AI development and says he doesn't want to cede edge to China

China’s Foreign Ministry said on Monday that all countries should work together on AI, while warning that “fearmongering, confrontation and vicious competition” would disrupt global AI governance.

Chinese officials have also rejected US allegations that Chinese AI developers engaged in aggressive efforts to extract capabilities from advanced American models through “distillation”. China’s Commerce Ministry called the claims groundless and accused Washington of pursuing a monopoly over the AI industry.

That puts Beijing and Amodei in an awkward position where he is calling for cooperation with China on frontier AI safety while simultaneously arguing that the US should restrict China's access to advanced chips and other AI capabilities.

The state-run Global Times went further, describing Amodei’s essay as a veiled call to contain China and, in essence, a “Cold War playbook” for the AI sector.

The disagreement reflects a deeper problem. Both Washington and Beijing want to prevent the other from establishing a technological chokepoint, while also seeking access to the other side’s AI advances.

That makes the politics of slowing AI considerably harder than the technology debate. Companies may want to slow down enough to make increasingly powerful systems safer. Governments, meanwhile, do not want to slow down so much that they lose the AI race.

And researchers are asking what if the race itself is becoming the danger?

That may explain why three of the industry’s most unlikely allies are suddenly agreeing that the AI race needs brakes. The fight now is over who gets to decide where those brakes should be.