Charles Mountbatten-Windsor is presently King of England. I accept this as a fact in much the same spirit as I accept that VAT is presently twenty per cent. Whether either arrangement is desirable is a separate question. Since he is presently the King, and since we are continually assured that the Monarchy is above politics, it is interesting that he has convened a gathering of politicians and some of the richest businessmen in the world to discuss how artificial intelligence should be controlled.

According to the BBC, the King recently summoned his gathering to Dumfries House in Scotland. Among those present were representatives of Nvidia, OpenAI and Anthropic, together with Google DeepMind’s “Sir” Demis Hassabis and the British Minister for Artificial Intelligence. The stated purpose was to consider whether a “shared set of principles” could guide the future use of artificial intelligence. The official account confirms that this was a gathering of technology companies, government and assorted representatives of what is now called “civil society”.

Mr Mountbatten-Windsor opened proceedings with a warning. Artificial intelligence, he said, might develop “darker capacities”, perhaps including the capacity “to take life”. There was, he continued, an urgent need to consider the “existential dangers” of the technology “falling into the wrong hands”. He wanted “sufficient means of control before it is all too late”.

Whenever I hear the expression “wrong hands”, my first question is: wrong according to whom? I will come to this. Before then, let us consider the alleged danger.

It is possible that artificial intelligence will destroy the human race. I have no means of knowing. Neither, I suspect, has Mr Mountbatten-Windsor. There are people who understand these matters better than either of us who take the danger seriously. Mr Hassabis told the meeting that artificial general intelligence might be only a few years away, and that the probability of disaster was not zero. Others have lately made much more alarming estimates. These warnings should not simply be laughed away.

At the same time, the human race already possesses technologies capable of killing on a scale beyond anything artificial intelligence has so far achieved. We have possessed them for eighty years. Nuclear weapons can destroy cities in minutes. Biological weapons could conceivably kill millions. Modern states possess missiles that can cross continents and submarines able to incinerate countries without warning. There has been no shortage of warnings about the existential danger of these things. What I do not remember is the governing classes of the world concluding that governments ought therefore not to possess them. They have worried incessantly that the weapons might fall into the “wrong hands”. Their principal concern has always been that they should remain in the right hands—which invariably means their own.

Nuclear weapons were politically reassuring because ordinary people could not have them. The British Government could build an atomic bomb. I could not. The Soviet Government could manufacture hydrogen bombs. A dissident in Leningrad could not make one in his kitchen. Nuclear weapons vastly increased the destructive power of states while doing nothing to increase the defensive power of their subjects. They therefore fitted perfectly into the political order of the twentieth century.

Indeed, the Bomb was almost an ideal establishment technology. It required immense concentrations of capital, scientific expertise, industrial plant and state secrecy. Its development strengthened governments, while its possession justified further secrecy and the enlargement of the national security apparatus. Whether it protected the population or threatened the population could be debated. What cannot be debated is that it strengthened the State against everyone who did not possess one.

Artificial intelligence is different. Its most politically important characteristic may turn out not to be that it can kill us, but that we can use it.

After it first became a consumer technology, I regarded the extravagant claims made for artificial intelligence with suspicion. Bryan Mercadente and other young enthusiasts could chatter about every new model as if Prometheus had once again come down from Olympus. I preferred to wait and see.

I have now seen enough to revise my opinion. I have been using artificial intelligence to deal with a series of disputes with my local planning department. The experience has not persuaded me that the Singularity is upon us, or that we shall presently upload ourselves into computers and live for ever. It has shown me something more immediately useful. Artificial intelligence has enabled me to meet the bureaucracy on something approaching equal terms.

For centuries, governments have possessed an overwhelming informational advantage over their subjects. This did not begin with computers. The Domesday Book was compiled in the eleventh century. Early modern states accumulated tax and parish records. By the nineteenth century, governments could conduct censuses and maintain immense paper archives. The twentieth-century totalitarian states achieved astonishing levels of surveillance with filing cabinets, index cards, telephones and human informers.

Information technology makes surveillance cheaper. Artificial intelligence may make it cheaper still. But this is a difference of degree. The State already had the ability to know an immense amount about us. The revolutionary change is in the opposite direction. We can now know about it.

Consider the normal relationship between an ordinary citizen and the British administrative state. The citizen receives a decision he dislikes. He is told that he can challenge it. The procedure is contained in a document he has never heard of. This document refers to statutory guidance, which refers in turn to regulations that have been amended several times. The council has its own published policy. There are relevant decisions by inspectors and perhaps by courts. There is also a complaints procedure, followed by another complaints procedure when the first complaints procedure has been exhausted.

The citizen has a job. He has a family. He would rather spend his evening reading a book or watching television. The official on the other side deals with this material every day. That inequality is not incidental to the administrative state. It is one of its foundations. The citizen theoretically possesses rights. The practical exercise of those rights requires either money or obsession. He can pay a solicitor or consultant several hundred pounds an hour. Alternatively, he can turn himself into one of those unfortunate men who spend twenty years fighting the council over a garden wall, until they possess fourteen filing cabinets, know every paragraph of the Town and Country Planning Act and have ceased receiving Christmas cards from their children. Most people choose surrender.

Artificial intelligence alters this arrangement. Give it the decision, the regulations, the correspondence and the council’s policy. Ask it what has not been answered and whether the stated procedure has been followed. Ask it to compare today’s explanation with the explanation given six months ago. Ask it to identify the relevant provisions and draft a response. Then check the citations yourself. AI can make mistakes, and anyone who submits an invented legal authority deserves the consequences. But most of the drudgery has disappeared. The British State spent decades making its procedures incomprehensible. We now have a machine that can read them. No wonder there is alarm.

Indeed, the alarm had become visible before Mr Mountbatten-Windsor assembled his grandees in Scotland. A few months ago, I discussed what has been called “agentic flooding”: the sudden growth of AI-assisted complaints, claims and applications that threatens to overwhelm public bodies. This is a revealing expression. What appears as a flood from the point of view of the bureaucracy may look rather different from the point of view of people who were previously deterred from exercising their supposed rights.

The fear is not merely that artificial intelligence will produce nonsense. It certainly can, though so can human beings and government departments. The greater difficulty will arise as AI becomes better at producing complaints and appeals that are accurately researched, competently framed and persistent. If everyone can have something approaching the services of a junior solicitor for the price of an electricity bill, the economics of administrative government begin to change.

There is something most revealing about this. For generations, Parliament and the administrative state have surrounded themselves with procedural safeguards. There are consultations and appeals, Freedom of Information rights, ombudsmen and tribunals. All these things are paraded as evidence that we live under accountable government. There was merely an unstated proviso. We were not supposed to use them too much. The system was designed in the comfortable expectation that most people would give up. Parliament could create a right because exercising it was expensive. The bureaucracy could tolerate an appeal because relatively few people possessed the time or knowledge to bring one. The State could proclaim its transparency because finding anything out remained an occupation in itself. Now ordinary people have acquired tireless clerks who understand bureaucratic English, never become bored and charge almost nothing. The response is that there may be too many complaints.

You might have expected a different response. If citizens can use artificial intelligence to reveal that administrative decisions are badly reasoned, perhaps administrative decisions should improve. If a tribunal cannot cope with the number of people asserting rights Parliament has granted them, perhaps the legislation should be simplified. If a planning system collapses when citizens understand it, perhaps the planning system deserves to collapse. Instead, we hear about “safety”.

This is where Mr Mountbatten-Windsor’s phrase “wrong hands” becomes interesting. Obviously, there are genuinely dangerous uses of artificial intelligence. A terrorist using AI to design some new means of killing people is not exercising a civil liberty I feel any obligation to defend. Nor should we pretend that sophisticated systems cannot create new dangers merely because governments are opportunistic in describing them. But observe the language that invariably surrounds demands for control. We shall hear about child protection and terrorism. We shall hear about extremism and the “far right”, followed inevitably by misinformation. Each concern may contain some truth. Together they establish the premise that artificial intelligence is too dangerous to remain substantially beyond political control.

We have seen this performance before. The administrative state rarely begins by saying that it wishes to protect itself. It discovers a victim. Children are particularly useful; terrorism is nearly as good. Once the principle of control has been conceded for an object no reasonable person can oppose, its boundaries can be expanded at leisure.

The Internet itself went through this process. First came understandable concerns about crime; presently the vocabulary expanded to “harmful content”, misinformation and disinformation. The words grew softer as the intended power grew harder. AI presents a greater challenge because it attacks the informational foundation of managerial government. No one understands the whole apparatus of British law, regulation and administrative discretion. That is part of its strength. The citizen wanders through it like Joseph K. through the offices of Kafka’s Trial, forever told that the answer exists somewhere else. Artificial intelligence can begin joining the pieces together.

A search engine can give me ten thousand documents and thereby increase my helplessness. An AI system can tell me which six matter, compare them, and notice where one contradicts another. For the first time, the individual can possess something resembling a bureaucracy of his own. The State still has the police, the courts and the tax collectors; but its monopoly of institutional memory is weakening. Printing weakened those who had prospered from the scarcity of books. The Internet weakened those who had prospered from scarcity of publication. Artificial intelligence may weaken those who prosper from scarcity of comprehension. That last monopoly is immensely valuable.

We should therefore look with suspicion on the strange alliance gathering around AI regulation. At Dumfries House we had the Crown and the Government sitting down with representatives of some of the largest technology corporations on earth to discuss principles for controlling a technology whose most interesting characteristic is its ability to confer specialist capacities on ordinary people. The official account speaks of “thoughtful stewardship” and technology being “carefully curated”. I dislike the word curated outside museums. Applied to technology, it means someone else deciding what I may have.

Mr Mountbatten-Windsor asks how artificial intelligence can be kept “firmly in the service of humanity”. I would answer that one promising way is to leave it in the hands of human beings. This does not mean governments or approved corporations operating under government licences. Nor does it mean accredited researchers, royal foundations and committees of ethical experts. It means human beings.

If artificial intelligence makes it easier to challenge a tax demand, fight a planning department, interrogate official statistics or discover that the regulator’s latest explanation contradicts the one it gave six months ago, so much the better. Tenants will use it against landlords, small employers against regulators, and defendants to understand accusations expressed in the professional dialect of those accusing them. There will be abuses, and there will certainly be inconvenience. I find the second prospect particularly cheering. Liberty is frequently inconvenient to those who exercise power.

I do not know whether artificial intelligence will eventually develop consciousness and decide to exterminate us. If that happens, I shall concede that Mr Mountbatten-Windsor had a point, though not perhaps for very long. The immediate danger is easier to see. It is that the governing class will use hypothetical catastrophe to secure control over an actually liberating technology. They were comfortable enough with technologies that allowed governments to destroy cities. They are becoming nervous about one that allows a householder to write a competent letter to the council.

The “danger” for Mr Mountbatten-Windsor and his friends is not that artificial intelligence may fall into the wrong hands. It is that it has fallen into ours.