Silicon, Sovereignty, and the State: The Geopolitical Brawl Over Frontier AI

On June 12, the world learned precisely what modern superpower leverage looks like. With a single directive from the Trump administration, artificial intelligence lab Anthropic shut out every non-American user from Fable 5 and Mythos 5, its latest and most formidable frontier models. The ban hit everyone, including foreign nationals on Anthropic’s own payroll, not just target geopolitical rivals. Over night, the American government established itself as the gatekeeper of the world's most critical technology.

The official justification was a "jailbreak," a user prompt that slipped past Fable’s safety guardrails, potentially allowing the model to assist in cyber-attacks or biological weapon design, yet the wider strategy is unmistakable: the administration likely wanted an instant emergency brake pulled on the models, and blocking foreigners was the quickest legal lever available. While the exact statutory grounding remains murky and a blanket ban won't last for ever, the message hit home: global access to cutting-edge cognitive power now depends on the mood in the Oval Office.

This heavy-handed intervention follows a pattern of escalating state control. Back in March, the administration designated Anthropic a "supply-chain risk" during a bitter public dispute over autonomous weaponry. Now, it has shown it is equally willing to disrupt the lives of millions of consumers and corporate clients worldwide. As allies and adversaries alike scramble to adapt to a transactional, unpredictable America, they must confront a new reality: being economically and technologically captive to the world's dominant AI hegemon.

Who wins when a Silicon Valley high priest battles an imperious president?

To understand how we reached this point, you have to look at the two men at the centre of the storm: Donald Trump, America’s imperious commander-in-chief, and Dario Amodei, the chief executive, or, as some industry insiders joke, the supreme pontiff, of Anthropic.

History shows that moral authority often derives from a refusal to engage in the dirty compromises of worldly politics. This dynamic makes spiritual leaders easy to bully in the short term, because they lack physical weapons, but remarkably resilient over time, as persecution only burnishes their reputation. Amodei has built Anthropic with a distinctly monastic ethos. Founded in 2021 by Dario, his sister Daniela, and six defectors from OpenAI who felt their former employer wasn't taking safety seriously enough, the San Francisco lab operates like a secular priesthood dedicated to the mystery of artificial general intelligence (AGI).

When Pope Leo XIV issued his landmark AI encyclical, Magnifica humanitas, an Anthropic co-founder stood at his shoulder. Amodei regularly pens extensive essays of his own on exponential risks and governance; his latest, Policy on the AI Exponential, dropped on June 10th with an urgent plea for binding federal regulation. Every fortnight, he delivers hour-long sermons on safety to his 3,000 employees, many of whom are famously "AGI-pilled," convinced they are building a machine that will eventually render human labour obsolete, yet morally bound to ensure it doesn't destroy humanity first.

Remarkably, all this ethical hand-wringing has been an absolute commercial goldmine. While OpenAI spent years capturing the public imagination with consumer chatbots, Anthropic took a quieter, more lucrative bet on corporate clients. Its coding assistant, Claude Code, became an overnight sensation among software engineers, executing multi-day coding tasks in hours with verifiable reasoning. Businesses, terrified by the prospect of rogue AI, flocked to a brand that made safety its main selling point.

The financial divergence is striking. Four-fifths of Anthropic’s revenue now comes from business customers, compared to roughly 40% at OpenAI. As both firms head towards initial public offerings (IPOs) expected to value each around $1trn, Anthropic's annual recurring revenue (ARR) has surged past its rival, growing fivefold in just five months. By May, PitchBook and Futurum data placed Anthropic’s valuation at $965bn against OpenAI’s $852bn in March, with Anthropic on track to post the industry's first operating profit in the second quarter. At roughly 20 times ARR, analysts like PitchBook's Harrison Rolfes note that while Anthropic isn't cheap, its growth rate makes the valuation surprisingly sane, especially compared to SpaceX, which traded at over 90 times revenue after its own 12 June listing surge.

Yet Anthropic's meteoric rise and regulatory lobbying deeply irritated Washington. Trump advisors, notably David Sacks, sneered at the lab’s "doomer industrial complex," accusing Amodei of running a sophisticated regulatory-capture scheme based on fearmongering to slow down competitors while quietly building ever more lethal models. Sacks even compared Anthropic’s exponential power grab to John D. Rockefeller’s 19th-century oil monopoly.

When Anthropic and the Pentagon publicly clashed in February over using Claude for mass domestic surveillance and autonomous killing machines, Trump branded the company's leadership "leftwing nut jobs." The administration realised that Antropic was a sovereign rival attempting to dictate national security policy, not just another tech vendor.

How do you lock down a technology that thinks for itself?

The underlying dilemma transcends personal animosity. High-powered AI has evolved from a commercial tool into a weapon of mass disruption. On June 11, Senate Intelligence Committee vice-chair Mark Warner revealed that General Joshua Rudd, commander of the National Security Agency and US Cyber Command, testified that Mythos 5 had systematically breached almost all of America's classified military networks "not in weeks, but in hours."

When Amazon, a major Anthropic shareholder, discovered it could "jailbreak" Fable 5's guardrails, Sacks took to social media to accuse Amodei of refusing to patch the vulnerabilities. Anthropic privately disputed this, noting they had given the government weeks to test the models before release and pointing out that foolproof AI security is mathematically impossible. But with no calibrated regulatory framework in place, Trump having scrapped previous rules in favour of a voluntary executive order issued on June 2, the White House resorted to the blunt instrument of export controls, treating Fable and Mythos like munitions.

We have seen this before. After the Second World War, despite Britain having shared its foundational atomic research, the US Congress passed the 1946 McMahon Act, abruptly cutting off all nuclear co-operation with foreign powers, allies included. It took years before Britain developed its own bomb and forced America back to the negotiating table. Similarly, during the crypto wars of the 1970s to 1990s, the US government treated public-key cryptography as a military weapon, investigating developers under International Traffic in Arms Regulations before civil-liberties campaigners forced a retreat. America still refuses to share its top-tier F-22 stealth fighters even with close friends, selling them the slightly downgraded F-35 instead.

AI, however, is infinitely more pervasive than encryption and far harder to contain than enriched uranium. The US currently holds a commanding lead, sitting roughly a year ahead of a Chinese tech sector hobbled by American chip sanctions. If an American lab cracks recursive self-improvement (RSI), where AI models begin autonomously coding superior versions of themselves, that lead could become permanent.

This dynamic is pushing Washington toward a strict global hierarchy of access. The supreme capabilities will be hoarded domestically for cyber-offence and military dominance. A second tier of "good enough" models (the F-35s of the AI world) will be leased to trusted allies. The rest of the globe will have to settle for heavily handicapped systems or open-weight alternatives. Yet as former FBI cyber official Cynthia Kaiser notes, prohibition is a leaky vessel; hackers are already buying American digital identities on the black market to bypass geographical blocks, just as Chinese coders regularly slip past Anthropic's regional IP restrictions. Furthermore, freezing out foreign researchers, who make up a massive chunk of American lab staffs, risks grinding US innovation to a halt.

What happens when America's closest allies are treated like potential adversaries?

For America's closest friends, the June 12 order was a profound diplomatic shock. Across Europe and the Five Eyes intelligence alliance (Australia, Britain, Canada, and New Zealand), government agencies, financial institutions, and tech corporations found their access to Mythos 5 revoked overnight. Even Britain’s AI Security Institute, widely regarded as the world's premier testing ground for jailbreaking frontier models, was suddenly left staring at an access-denied screen. Former British security minister Tom Tugendhat captured the mood bluntly: "After a lesson this clear, every nation will be asking what they need to achieve sovereignty."

The economic blast radius extends far beyond NATO. Consider emerging digital hubs like Metro Manila, whose sprawling Business Process Outsourcing (BPO) and IT-services sector is undergoing a massive structural shift. The Philippine economy relies heavily on integrating enterprise tools like Claude Code to upgrade from basic customer support to high-value software engineering, legal analysis, and financial diagnostics. If the US government decides that non-American workers cannot access frontier cognitive tools, or if developing tech hubs are permanently relegated to slow, nerfed AI models, entire service economies risk losing their global competitiveness overnight. That's like trying to compete in Formula 1 but the FIA only gave you a go-kart.

While British and European spy agencies are already quietly negotiating to get their Mythos credentials restored, as former officials confirm, the broader dependency remains terrifying. Geopolitical analysts like Anton Leicht at the Carnegie Endowment point out the brutal reality for middle powers: a Trump administration that refuses to let you access Fable is hardly going to sell you enough advanced Nvidia chips to train an unrestricted Fable clone yourself. Europe currently hosts a mere 5% of global AI compute infrastructure, while America controls over 75%. If Elon Musk's SpaceX succeeds in deploying orbital data centres, that compute gap could become insurmountable, turning allies into permanent digital vassals.

Can the rest of the world build a way out of digital vassalage?

Resignation, however, is not a strategy, nor is retreating into protectionism. As tech researcher Judith Dada argues, the only viable response to America’s warning shot is the most ambitious peacetime political and industrial agenda ever conceived. To stop whining and start building leverage, the global community, led by Europe and allied tech nations, must execute three urgent policy pivots.

First, middle powers must form a strategic AI cartel to aggregate their scattered leverage. No single nation can bully Washington alone, but together they control the physical bottlenecks of the digital age. The Netherlands houses ASML, the undisputed monopoly in extreme ultraviolet lithography machines required to make advanced chips. South Korea dominates high-bandwidth memory production. Japan controls vital semiconductor-manufacturing equipment. Norway, France, and Canada sit on massive reserves of clean energy, while Germany holds deep industrial data and Britain offers world-leading AI security talent. By combining these assets, an allied coalition could force the US into legally binding access treaties: American labs get the machinery, memory, and energy they desperately need, but only in exchange for unrestricted, sovereign access to frontier models.

Second, the rest of the world must radically scale up its compute capacity. For Europe, this means setting a hard target to jump from 5% to 20% of global AI compute within five years. That requires building tens of gigawatts of dedicated infrastructure, a logistical mountain that cannot be scaled through old-school bureaucratic dawdling. The scale of the energy challenge is staggering; in Britain alone, grid connection requests surged from 41GW to 125GW in just seven months leading up to June 2025, forcing OpenAI to pause a major UK data centre in April citing red tape and extortionate power costs. Governments must establish specialised economic zones with fast-tracked planning permissions and massive energy generation investments, embracing "sovereignty-as-interdependence" by inviting US tech giants to build facilities on foreign soil under shared-governance rules.

Third, structural market reforms are essential to stop economic drag. The European Union's traditional strength (its massive consumer market) is being eroded by rigid labour laws that penalise companies for adopting efficiency-boosting technologies. To survive the AI transition, nations must pivot toward Denmark's "flexicurity" model. This approach allows businesses to swiftly adopt transformative AI and restructure their workforces, while the state provides displaced workers with robust income support and mandatory, high-level retraining. Freezing jobs in place today simply ensures that entire companies will be wiped out by faster-moving American or Asian competitors tomorrow.

Ultimately, however, legislation and data centres are just infrastructure; they don't answer the human question. Why should ordinary citizens accept years of disruption, construction noise, and shifting job markets just to build AI servers? As Dada rightly notes, people do not adopt technology simply for the sake of it. The hardest challenge facing political leaders is crafting a compelling, positive narrative that unites society behind this transformation.

The story cannot simply be about preventing American dominance or surviving workplace automation. It must be a vision of how sovereign, ethical AI can actively improve human lives, revolutionising healthcare, accelerating scientific discovery, and creating entirely new avenues for creative and economic ambition, all while maintaining strict safety standards that an unchecked, transactional superpower might ignore.

The June ban on Anthropic will eventually be litigated, softened, or withdrawn as commercial realities set in. But the illusion of open, borderless AI is dead. The tools that will define the next century of human progress are now instruments of state power. The rest of the world must decide right now whether to build the leverage required to shape its own destiny, or spend the future waiting for permission to log on.