The news that Anthropic has had to take Fable 5 and Mythos 5 offline for foreign users because of US export restrictions should send a shockwave through everyone involved in AI in the UK.
This is a strategic dependency warning. According to AP, Anthropic said it disabled access to its latest models after a US government directive aimed at preventing use by foreign nationals. Business Insider reported that the order covered Fable 5 and Mythos 5 and even affected some foreign-born Anthropic staff.
Whatever the duration of that decision, the UK has just been given a live demonstration of dependency risk. If the most important models in the world are controlled elsewhere, then our access to them exists at the pleasure of someone else's policy, someone else's security establishment, and someone else's election cycle.
We urgently need to develop our own sovereign frontier models: models that are safe, competitive, and capable of ensuring the UK remains relevant in this space.
At the moment, the environment in the UK prevents this at the scale required. Energy prices are often cited as one of the major factors, and rightly so. Energy is only one piece of the puzzle. We also have poor rates of pay for British talent compared with the US, which gives people little incentive to stay here once they have the skills the market wants.
On top of that, taxation is eye-watering for high earners. For top talent, there is often no compelling reason to remain in the UK beyond personal ties. A country cannot build strategy out of sentimentality with a Treasury receipt stapled to it.
The Lesson From Fable 5
The Fable 5 incident matters because it makes a theoretical problem painfully concrete. For years, "sovereign AI" has sounded like a useful, bloodless think-tank phrase. Now it has a very simple meaning: can British researchers, companies, public bodies, and national security teams use the best models when they need them?
If the answer is "only when Washington says yes", we have a problem.
The UK should keep deep partnerships with Anthropic, OpenAI, Google, Meta, Microsoft and the rest, while building enough domestic capacity to avoid dependence. A country that cannot train, host, evaluate, and improve frontier-scale systems on its own soil remains a customer.
The Government knows this, at least on paper. Its AI Opportunities Action Plan response commits to expanding sovereign compute capacity by at least 20x by 2030, creating AI Growth Zones with enhanced access to power, and launching a function to strengthen the UK's sovereign AI capabilities.
Those are sensible moves. They are still too slow and too narrow for the moment we are in.
Compute Is An Energy Policy
Frontier AI is software plus energy, land, grid connections, cooling, hardware supply, planning permission, and people. If any one of those is weak, the whole stack wobbles.
The UK has a particular problem with power. Large training runs and serious inference capacity need cheap, reliable electricity at industrial scale. A country that makes power expensive is effectively taxing intelligence infrastructure before it even gets out of the ground.
Recent comparisons look grim. The Times reported that British industry paid around £168/MWh for electricity in 2024, compared with £69/MWh for France and £38/MWh in Texas. The US Energy Information Administration puts the 2024 average US industrial electricity price at 8.13 cents/kWh. Eurostat shows how wide the spread is inside Europe, with non-household electricity in the second half of 2025 far cheaper in Finland than in Germany or Ireland.
The details vary by contract, user size, tax treatment and time period. The signal is obvious enough: AI infrastructure will follow power. The UK cannot become a frontier AI country while treating industrial electricity as an afterthought.
The Government needs to rise to the moment and ensure our energy policy is fit for the future, with a clean break from the ill-conceived set of policies created in the early 2010s. Things have moved on. AI data centres are becoming part of the industrial base.
Talent Will Go Where It Is Rewarded
Even if the UK fixes compute, it still needs the people.
This is where our national conversation gets weirdly coy. We talk about skills, fellowships, scholarships, visas and clusters. Fine. We need all of that. But elite technical people also respond to pay, ownership and tax.
British AI talent is world-class. The problem is that world-class talent can leave. The US offers higher salaries, deeper capital markets, more aggressive equity packages, and a tax system that, while hardly simple, avoids the same bizarre punishment zone we impose around £100k.
In England, Wales and Northern Ireland, the current UK income tax system removes the personal allowance by £1 for every £2 of adjusted net income above £100,000, until it disappears entirely at £125,140. Add employee National Insurance, and the marginal employee deduction in that band is about 62%.
That is an absurd signal to send to exactly the people we claim we want to keep.
The US can still be painful once state taxes are included, especially in California and New York. The federal baseline is lower for many high earners, and the US also contains zero-income-tax states. The Social Security wage base caps the 6.2% employee Social Security tax at $184,500 in 2026, while Medicare continues at 1.45% with an extra 0.9% above $200,000. The 2026 federal income tax brackets leave a single filer on $250k with a federal-plus-FICA burden of roughly 26.7% before state tax.
The UK equivalent employee on £150k loses about 39.1% to income tax and employee NI before pension choices, student loans, childcare cliffs or anything else. More importantly, the marginal trap starts much lower.
For founders, researchers and senior engineers, this matters. People can do maths. They can compare compensation. They can read a term sheet. If the UK wants them to build here, it has to stop behaving as though personal attachment to Britain will compensate indefinitely for worse pay, higher effective tax, and expensive infrastructure.
Token Gestures Are Not Enough
The Government urgently needs to focus on the logistical side of this challenge and offer more than the token gestures we have seen so far.
That should include serious tax reform for strategic sectors such as AI. Actual reform that makes the UK a rational place to build a frontier AI company, join one early, and stay long enough for it to matter.
There are many ways to do that. A targeted treatment for equity compensation in strategic technology companies. A removal of the £100k personal allowance taper. A clearer regime for researchers and founders relocating to the UK. More generous treatment for deep-tech employee ownership. A tax credit that follows high-value technical labour and avoids consultancy theatre.
Some of these ideas will be imperfect. All of them are more serious than pretending that a scholarship programme and a press release will compete with US labs offering huge salaries, deep compute, and a credible path to global scale.
Energy policy needs the same seriousness. If AI Growth Zones are going to work, they need real power allocations, faster grid connections, planning speed, and a route to long-term cheap electricity. Nuclear, renewables, storage, grid reform, demand flexibility, and direct industrial power deals all need to be on the table. The right answer is abundant power.
A Chance To Come In From The Cold
This is one of the clearest opportunities for the UK to come in from the cold and become relevant again.
We have advantages. We have extraordinary universities. We have DeepMind's legacy, a strong safety and evaluations ecosystem, serious financial markets, good startup law, and a language and timezone position that still matters. We have enough technical credibility for Britain to make a serious claim in AI.
Credibility expires when left unused.
The UK should aim higher than becoming the place where US companies open London offices, hire our best people, and route the strategic value back across the Atlantic. We should also aim higher than being a regulator, evaluator, buyer, or polite conference host. Those roles matter. They fall short of sovereignty.
Sovereignty means having enough domestic capability to keep access in British hands. It means compute onshore. It means energy policy that understands the next industrial base. It means tax policy that rewards building here. It means companies with the capital, talent and ambition to train models that the rest of the world has to take seriously.
The Fable 5 shock should be treated as a warning shot. The UK can either respond like a country that intends to shape the next technological era, or it can continue issuing strategies while the infrastructure, money and people move elsewhere.
Right now, we are squandering it.
