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

The bosses of the world's leading AI companies have a clear ask for the leaders of the G7: you have to come up with a way to govern artificial intelligence.

But they disagree sharply on exactly what that should mean.

This was the message inside the room as the leaders of the G7 wrapped up three days of talks in the French Alps with a working lunch on the fraught topic of AI.

Joining them were the heads of the three leading AI companies: OpenAI CEO Sam Altman, Google DeepMind chief Demis Hassabis and Anthropic boss Dario Amodei.

The conversation came at a tense time.

Less than a week ago, the Trump administration forced Anthropic to cut off access to its most advanced AI model, igniting concerns among Western nations about the "kill switch" Washington has over digital technology.

Yet, according to the people familiar with the talks, no-one mentioned the incident directly.

Instead, the three AI bosses focused their remarks on themes of collaboration and regulation, making their pitch for the future of AI.

Some not-so-reassuring thoughts...

Hassabis, Altman and Amodei warned the assembled leaders that they needed to act fast. The exact timelines varied, but all agreed: governments have very little time to get a grip on AI.

Hassabis talked about a three- to five-year timeline. Amodei said we were one to two years away from AI models being better than humans at everything. Altman said: "In another year or two, I expect we will have built systems with astonishing power."

Yet although all three emphasised the immense benefits this would bring, they expressed open concern about the downsides.

Not jobs or inequality – they were barely mentioned. These were national security concerns: cyber, bioterrorism, nuclear and warfare.

If the current trajectory continues, Amodei warned, AI could be the dominant source of economic and military power for nations.

Not a reassuring thought for the assembled leaders outside the US, where all the leading AI companies are ultimately based (DeepMind is in London but is owned by Google).

Faced with this prospect, all three AI bosses spoke of the need for international governance. Hassabis said the issue was too important to be left to technologists, saying "we need to define this together." Altman was even firmer. "Do not cede your responsibilities to AI labs like mine," he told the G7 leaders.

But the AI bosses diverged sharply on who should control access, how to secure the technology, and how much authority AI labs should have.

Liberty or control?

For instance, Amodei called for a US-led coalition of democratic countries to control access to the technology and "isolate common adversaries" such as China.

Altman rejected this, saying that once guardrails were in place, "we must err toward human liberty. We want everyone on earth to benefit from this technology and to figure out for themselves how to use it".

Another faultline was over the role of the AI labs.

Hassabis called for "a body that increases trust", suggesting a "technical standards body that is supported by leading labs". This new body – international but led by the US – could adapt its standards every quarter as new risks became clear, something Hassabis said he had been working on.

Altman, by contrast, warned that the AI labs should not have too much control.

"There is a threat more insidious than the technical risks of this technology," he added. "It is the threat that the very real risks that AI poses become the justification for concentrating power in the hands of the few."

Read more from our experts:
Trump's war has been a waste of time
America's Iran deal is an admission of defeat

The crucial question: should frontier AI governance be built around containment or broad access? And should it be experts or national governments that decide?

In his remarks, Sir Keir Starmer did not answer that question directly, choosing instead to focus on public consent.

The UK prime minister agreed that standards were urgent given the pace of AI progress, saying: "If AI leaders are worried about this, politicians are worried."

But rather than setting out his position on the best framework for international governance, he called for greater understanding of the need for public consent.

It was vital to be able to protect children and free speech and advance AI, he said.

"If we can't do both things," he told the room, "we will lose the consent of the public."

(c) Sky News 2026: AI's most powerful bosses deliver message to world leaders - and it's not very reassuring