Generated by Codex with GPT-5

This summary covers The Economist’s June 20th, 2026 Leaders article listed in the contents under Artificial intelligence and published under the headline America's AI power grab.

The article argues that America’s most important new geopolitical asset may not be a carrier group, a reserve currency or a trade route. It may be the ability to decide who can use the most capable frontier AI models. When the Trump administration ordered Anthropic to block non-Americans from its latest Fable and Mythos models, the immediate trigger was a reported jailbreak. The broader message was larger: Washington can turn access to the world’s leading AI systems into an instrument of state power.

The order may not survive for long, and its legal basis remains uncertain. That is almost beside the point. Anthropic reportedly could not separate American from foreign users cleanly, so access was suspended more widely. In one move, a private AI lab, foreign customers and even non-American employees learned that the American government can intervene directly in the availability of frontier models. The Economist treats that as a demonstration of leverage, not just a temporary safety dispute.

A new kind of strategic choke point

The article places frontier AI in a line of earlier technologies that America tried to control. After the second world war, it withheld nuclear-weapons assistance from Britain until Britain developed enough capability of its own. In the early decades of modern cryptography, America restricted exports before accepting that secrecy could not fully contain the field. It still keeps some military technology, such as its most advanced aircraft, for itself while offering allies a less capable version.

AI resembles each case without fitting any of them neatly. If the best models can help users discover cyber vulnerabilities, design biological threats or operate military systems, then unfettered public access looks dangerous. But if the underlying methods leak, if open-weight models keep improving or if rivals train their own systems, control will be partial and temporary. The technology is both a security risk and a general-purpose economic engine.

That dual nature creates a strategic dilemma for America. It has strong reasons to hold back the most dangerous capabilities, especially where cyberwarfare and military systems are concerned. It also has strong reasons to sell AI services abroad. Foreign users are a large part of the market. Foreign researchers are part of the talent base inside American labs. Allies that feel permanently excluded may try to reduce dependence on American suppliers or lean more heavily toward China, the next-largest AI power.

The likely result is tiered access

The Economist’s central forecast is a hierarchy. America may reserve the most capable systems for itself, make somewhat weaker versions available to trusted allies and sell more restricted models to the broader world. This would echo the way America handles sensitive military exports, but with a technology that could sit inside office software, cybersecurity tools, weapons systems, scientific research and industrial planning.

For allies, that prospect is uncomfortable. They already face a more transactional America on defence, trade and the dollar system. AI could become another point of dependence, and perhaps the most consequential one. Europe, in particular, is far behind the leading American labs. The article notes that running frontier models requires immense computing power, where America has a large lead and is still investing at a much faster pace.

Dependence is not one-way. America still needs Dutch lithography machines, Taiwanese semiconductors and allied markets. Foreign governments and companies could also use good-enough open models if American tools become too politically risky. But those alternatives do not erase the main problem. If the largest clusters, the strongest models and the deepest AI companies remain concentrated in America, other rich democracies will negotiate from weakness.

What others can still do

The article’s advice is blunt: allies should stop treating AI capacity as a normal industrial-policy trophy and start treating it as strategic infrastructure. That means more power generation, faster grid connections, easier permitting for data centres and regulatory systems that do not drive frontier companies elsewhere. Britain is given as a warning case, with grid-connection demand rising far beyond peak electricity use and OpenAI pausing a data-centre project because of energy and regulatory constraints.

Europe is not the only region with work to do. In East Asia, Taiwan, Japan and South Korea have complementary strengths, but the article argues that they should integrate their efforts rather than duplicating them. The goal should not be autarky or protectionism. It should be enough local capability, compute and talent to make American AI power negotiable rather than absolute.

The closing lesson is pragmatic. If Washington is becoming more willing to use private technology as geopolitical leverage, complaining about alliances will not be enough. Countries that want autonomy need to build the physical and institutional base for AI: energy, compute, permissive planning, competitive labs and regional cooperation. The countries that fail to do so may discover that access to intelligence itself is something another capital can ration.