Here is something that should make you uncomfortable: text composed by a large language model has already made its way into an act of the British parliament. Not hypothetically. Not in a pilot program. Into law.

That is the central revelation in a sweeping New Statesman investigation published this week. The piece, built on dozens of interviews with current and former government officials, advisers, technologists, and MPs, paints a picture of a quiet power transfer happening inside the British state. Foreign-built AI models are now drafting emails between officials, summarising ministerial briefings, composing speeches delivered in the House of Commons, and influencing the Bank of England’s interest rate decisions.

And most of the people who are supposed to be governing the country have no idea how deep it goes.

The Sovereignty Problem Nobody Wants to Talk About

This is not a story about whether AI is good or bad. It is a story about power. Every foundational AI model used in the UK government was built in either the United States or China. That means the software shaping British policy decisions, defence analysis, and legislative drafting was created by companies answering to foreign interests, foreign investors, and foreign regulatory frameworks.

When a researcher met with a group of MPs in 2021 to warn them that AI models could be imbued with someone else’s politics, one of the elected representatives asked: "What’s AI?" That was five years ago. The adoption has only accelerated since.

Every student at Oxford is now being educated with the help of OpenAI. The BBC uses AI to redraft articles. Demis Hassabis personally briefed Starmer’s entire cabinet in November 2024, selling them a vision where LLMs would provide economic growth and a faster, more productive government by replacing the administrative work of civil servants. The Tony Blair Institute projected savings of £37 billion a year.

The pitch landed. The implementation is now everywhere. The oversight is almost nowhere.

The War Nobody Sees

A former senior government technology adviser described what is happening as a "war" between three factions: the engineers building AI models, the plutocrats who own them, and the politicians who think they control them. The politicians are losing.

Large language models are not calculators. They are persuasion engines. They are built to produce answers that will be accepted by users, not to compute truth. When a tool like that is embedded in the machinery of government, the distinction between "advising" and "deciding" starts to blur. And when the tool was built by Sam Altman’s company or a Chinese state-affiliated lab, the question of whose values are encoded in those outputs becomes genuinely urgent.

The investigation draws a clear line: this is not about whether AI is useful. It is about whether a country that does not build its own foundational AI models can claim genuine sovereignty when those models are writing its laws, briefing its ministers, and shaping its intelligence analysis.

Why This Matters Beyond Britain

Britain is not unique. It is just the first country to get caught. Every government that has adopted American or Chinese AI tools without building domestic alternatives faces the same vulnerability. Australia, Canada, Japan, the EU: all of them are feeding sensitive government operations through models they did not build, cannot audit, and do not fully understand.

The New Statesman investigation should be read by every head of state who has signed an enterprise deal with OpenAI or Google. Because the question it asks is simple: if the software writing your laws was built by a foreign company that answers to foreign shareholders, how sovereign are you really?

The answer, based on what is happening inside Westminster right now, is: less than you think.