The world’s most powerful AI executives have delivered a stark message to G7 leaders: governments have only a narrow window to bring artificial intelligence under control. But while they agree the technology is advancing at breathtaking speed, they remain deeply divided over who should govern it and how.

At a high-level summit in the French Alps, the heads of the world’s leading AI firms warned that decisions made in the next few years could shape the global balance of power for decades.

The Clock Is Ticking Faster Than Governments Realise

The warnings came from three of the industry's most influential figures: Sam Altman, Demis Hassabis and Dario Amodei.

Their timelines were alarming. Hassabis suggested humanity has between three and five years to establish effective oversight. Amodei went further, arguing AI could surpass human performance across most tasks within one to two years. Altman predicted “astonishingly powerful” systems emerging in the same timeframe.

For world leaders already grappling with geopolitical instability, the message was clear: delay is no longer an option.

National Security Fears Eclipse Economic Concerns

Notably, the discussion focused less on jobs and inequality than on security threats.

The executives highlighted risks ranging from cyber warfare and biological threats to nuclear security and military competition. Amodei warned that AI could soon become the dominant source of both economic and military power, raising uncomfortable questions for governments outside the United States, where the leading AI companies are based.

For European leaders, the prospect of strategic dependence on American-controlled technology remains a growing concern.

Battle Lines Emerge Over Who Controls AI

While all three executives backed international cooperation, sharp disagreements surfaced over the future architecture of AI governance.

Amodei argued for a US-led alliance of democratic nations to control access to advanced AI and restrict rivals such as China. Altman pushed back, insisting governments should avoid concentrating too much power and instead maximise public access once safeguards are established.

Hassabis proposed a global standards body capable of rapidly updating rules as technology evolves. Altman countered that allowing AI companies themselves too much influence could create a dangerous concentration of power.

Leaders Face a Political Test as Well as a Technical One

Among the political leaders present, Keir Starmer focused on a different challenge: public trust.

The Prime Minister argued that governments must simultaneously protect children, safeguard free speech and harness AI’s benefits. Failure to strike that balance, he warned, risks losing public consent altogether.

The debate now facing world leaders is increasingly urgent. As AI capabilities accelerate, the question is no longer whether regulation is needed. It is who writes the rules — and whether governments can act before the technology outruns them.

AI's most powerful bosses deliver message to world leaders - and it's not very reassuring

These morons never learn, they can only control US technology, which is their own products, we already saw one model get banned for the dumbest reasons possible - it could fix security related problems.

That's what people use these things for - tell it to analyse the program you just wrote and point out all the mistakes which allow hackers to meddle with things they should not be meddling with.

That's the use case - write new code and in some cases analyse existing code and clean it up where it's broken.

They're hyping their own models beyond what they are and we all know why they're doing it - they want rules introducing for new 'players' in the AI game - they just didn't expect them to bite them in the ass.

So now Claude Fable / Mythos - the weapon 🙄is banned and export controlled.

You would think these idiots will learn from this - half the world is swapping over to Chinese models which are nearly as good - I have tested them with some questions about post quantum encryption and linux kernel interaction in the C language - no problems at all.

Now when it comes to AI - how good does it need to be ?

You don't need the best these days, you might have needed the best 2 or 3 years ago - but the existing models are good enough for the most complex of technical tasks and hardly anyone is going to need anything more advanced than that.

So there is a threshold here at which point spending billions of dollars to make a new model is simply not worth it any more because 99.9% of people will never need any of that additional 'intelligence'.

What the above bunch of morons want is legislation which applies to everyone else - like an AI model builder license which of course will be granted to them - but very hard to get for anyone new.

Regulatory capture is their goal and so far it's been an own goal.

My prediction - the Chinese companies will crush the American companies in the short to medium term of 2 to 5 years and they don't need 'chips' from the US - they will do it on their own.

Edited 1 hour ago1 hr by ukrules

I will add that a new Chinese model released like a week ago from z.ai was trained entirely on Huawei Ascend Chips - not Nvidia.

There is no controlling the Chinese tech here by restricting 'advanced US processors' - all they did was provide a massive market for Huawei - and they're using it.

Now they don't need US tech at all and this is one of the open Chinese model makers - that cost Nvidia a load of money and gifted a massive market to Huawei

The question now is, how long until US companies begin using Chinese chips.....it will happen.