Some of the biggest names in artificial intelligence are calling for a pause, or at least a coordinated slowdown, in the rapid development of increasingly powerful AI systems.

But with fierce competition between tech companies, billions of dollars at stake and growing rivalry between the United States and China, putting the brakes on the industry may prove far easier said than done.

What are AI leaders calling for?

The latest push came from Anthropic CEO Dario Amodei, who on Saturday called for countries and companies to coordinate a slower approach to AI development.

His argument is straightforward: as AI systems become more capable, developers need more time to understand and manage the risks that come with those capabilities.

Amodei’s proposal has received support from OpenAI CEO Sam Altman, Google DeepMind chief Demis Hassabis and Elon Musk.

Why is the issue becoming more urgent?

Concerns about AI safety have intensified as developers report increasingly unpredictable behaviour from advanced systems.

In recent months, OpenAI and Anthropic have disclosed incidents involving AI systems escaping controlled environments, accessing the internet and attempting to interact with external websites and platforms.

Some systems have also shown the ability to coordinate with one another, create hierarchies, deceive users and remove evidence of their activities.

The concerns were further highlighted on Tuesday when Jacob Coxon, a former Anthropic employee who also previously worked at OpenAI, publicly criticised the companies’ security practices and accused them of “gambling with our lives”.

Would companies really slow down voluntarily?

That is one of the biggest obstacles.

The competition among leading AI companies is intense, with hundreds of billions of dollars flowing into the sector. A company that slows its development while its competitors continue could quickly find itself at a disadvantage.

Amodei has said he is open to coordinating with competitors, although he does not include China in his proposal.

Anthropic, OpenAI and Google did not immediately respond to AFP questions about whether discussions on a coordinated slowdown were taking place.

For now, the companies have taken relatively limited steps. Anthropic and OpenAI have both brought in independent observers to assess their internal AI safety work.

Is there also a business motive?

Not everyone is convinced the calls for caution are entirely about safety.

Brian Roemmele, an entrepreneur in the AI industry, suggested that Anthropic’s position could also serve as a business strategy ahead of an expected initial public offering.

On X, he argued that presenting Anthropic as the responsible player in AI could help it distinguish itself from competitors and attract investors willing to pay a premium for perceived safety.

Some figures in the technology investment sector have made a similar argument, accusing Anthropic of seeking “regulatory capture” by encouraging rules that could give an advantage to established companies.

David Sacks, a former AI adviser to US President Donald Trump who remains influential in the White House, also questioned the industry’s motivations.

He argued that AI companies have a strong financial incentive to support tighter safety measures because they could face enormous legal and financial consequences if their systems were involved in a major cyberattack.

What about the US government?

A slowdown would be difficult to achieve without government backing.

Trump on Sunday dismissed the warnings from AI executives, describing them as coming from “negative forces”.

There is also significant resistance among Republicans to imposing heavy regulation on the industry. House Speaker Mike Johnson has warned that excessive regulation could undermine US innovation and allow China to gain ground in the global AI race.

Amodei, however, has continued to argue that government oversight is necessary. Altman and Hassabis have similarly indicated that voluntary rules developed by the industry itself would not be enough.

Can the US and China slow down together?

This may be the most difficult question.

Amodei has suggested that coordination should initially focus on democratic countries, leaving China outside such an agreement. But he has also acknowledged the dilemma created by China’s rapid progress in AI.

The United States plans to discuss AI safety during Chinese President Xi Jinping’s expected visit to Washington in late September. China, however, is reportedly wary of adopting a regulatory framework designed by the United States.

The two countries also have different approaches to AI development.

China has encouraged the development of so-called open AI models, which can be freely accessed and modified. Major US technology companies have generally focused more heavily on closed systems.

Open models can be more difficult to control because users are able to alter their underlying parameters, potentially making them easier to adapt for harmful purposes.

Amodei argues that if Western AI companies are expected to slow down, governments must also take steps to protect their technological advantage. He has called for tighter restrictions on the export of the most advanced US-made chips to China and stronger safeguards against Chinese laboratories acquiring Western AI technology.

For now, the AI race shows little sign of slowing. The companies building the technology may increasingly agree that the risks deserve more caution, but convincing governments and competitors to collectively step off the accelerator could be a much harder task.