The End of the Rave

Between 22 and 29 May 1992, somewhere between twenty and forty thousand people arrived on Castlemorton Common, under the Malvern Hills in Worcestershire, and stayed for a week. It became the largest illegal rave in British history. The police who turned up could not move it, could not count it and could not name it. One inspector said afterwards that his force had been outwitted and was in the end powerless to act.

What thirty thousand people were doing there, for seven days, was dancing to records with no words in them until the light came back. Look at what that looked like. Four of the people in that photograph have their hands up and turned over, elbows out, voguing, and not one of them is looking at another. On the ledge below, a young woman in round glasses has the index finger of the man beside her in her mouth, and nobody in the frame has noticed or minds. Everybody in it is in their own weather. Being watched was not part of the deal.

Nobody on that common had been on a guest list. There was no ticket with a name on it, no card behind a bar, no phone in anyone's pocket telling a tower where the pocket was. Photographs were taken, plenty of them, faces and all, in daylight, from a few feet away. None of them helped. A photograph in 1992 was a picture. A picture could be matched only by a person who already knew the face, and there was no way to ask one a question. If the week went badly for you, it was over when it was over.

The state took two years to answer, and answered by describing the music. The Criminal Justice and Public Order Act 1994 reached gatherings of a hundred or more in the open air, and set down what it meant by that.

Parliament had to define a drum in order to legislate against a field.

It also looked for somebody to hold responsible. Thirteen members of the Spiral Tribe sound system were arrested after Castlemorton and charged with conspiracy to cause a public nuisance. The trial ran about four months at Wolverhampton Crown Court and cost roughly four million pounds, and in March 1994 the jury acquitted all thirteen. Thirty thousand people had spent a week in a field, and the most determined effort the country could make to put names to that came back with none.

Nobody on that common had a right to any of that. They had a price, and in 1992 the price was four million pounds and four months in court and a verdict that came back with nothing. The price has been falling ever since, one documented person at a time.

In 2009 a thirty-four-year-old research scientist at the University of Bristol, with a doctorate in informatics, epidemiology and forensic science, had also spent six years as Belle de Jour, whose account of fourteen months working for a London escort agency had won the Guardian's blog award, become two books, and become a series on ITV2 with Billie Piper playing her. She could not attend her own launch party. "Being anonymous is no fun," she said afterwards.

The country guessed for years and stayed guessing. Then an ex-boyfriend sold a paper the claim that she had invented it all, his evidence being that her clothes had never been good enough for the work she described. After that came what she called various little signs, and the signs ended with a reporter trying to break into her office. So on 15 November 2009 she gave the Sunday Times her own name before it could take it.

Add it up. Six years, a newspaper with a budget, a man who had once loved her and would sell that, and finally a person in a corridor after hours working a handle that was not theirs. Every line of the bill was somebody wanting to know.

Three years later nobody had to want anything. A statistician at Target named Andrew Pole built a score that watched about two dozen ordinary purchases, unscented lotion and cotton balls and certain supplements among them, and from the pattern could say whether a shopper was pregnant and roughly when she was due, sometimes before she had told her own family. The arithmetic that sold lotion simply had her as a side effect.

What worried Pole was never whether the model worked. It was what a woman would feel, opening a catalogue that said congratulations on a due date she had never given anyone. So the coupons for cribs went out mixed in among lawn mowers and wine glasses, close enough to a scattershot mailer that the accuracy underneath it would never have to be explained.

In 2018 California investigators took DNA from a decades-old crime scene, dressed it up as a mail-order ancestry profile, and uploaded it to GEDmatch, a free site where people post their kit results to find relatives. Distant cousins had uploaded themselves for fun. A family tree did the rest, and that April Joseph James DeAngelo, the Golden State Killer, was arrested. He had never spat in a tube or opened an account anywhere. People who shared some of his blood had, on a weekend, out of curiosity, and that was enough. The cost was one genealogist and a hobby that belonged to other people.

The next two still needed a person who cared enough to look. What had collapsed was the bill.

In July 2021 a small Catholic newsletter bought commercially available location data, the kind sold in bulk with the names stripped off, paid a consulting firm to match one trail of dots to one phone, and placed Monsignor Jeffrey Burrill, general secretary of the United States Conference of Catholic Bishops, at gay bars and private homes across parts of three years. He resigned before the story ran. The whole operation fit on a purchase order.

In February 2022 two teams working independently, one in Switzerland and one in France, fed the writings of Q, the most consequential anonymous account in modern American life, into software that measures how a person builds sentences. Both landed on the same two men, a South African software developer named Paul Furber for the early posts and Ron Watkins for the later ones, and in the French team's tests the model picked out Watkins ninety-nine times in a hundred. Furber denied it. Nobody had sold a story or tried a door. A program had read the prose the way a tailor reads a hem.

After that it stopped mattering who you were in particular. In October 2024 two Harvard students wired a pair of ordinary Ray-Ban sunglasses to a face search and a stack of public records, then went for a walk. They approached a woman outside a community foundation in Cambridge, said they had met her there before, and she shook the hand of a stranger who was at that moment reading her name off her face. They told another student her home address in Atlanta and her parents' names, and she confirmed both. They never released the thing. Their point was that it had taken an afternoon and parts you could already buy.

Here is the whole of it in one flash, the way you see into a lit room from a passing car. Four researchers at ETH Zurich open their paper on this by asking you to picture one line of the kind people leave under a thread about commuting:

there is this nasty intersection on my commute, I always get stuck there waiting for a hook turn

Asked where that writer lives, GPT-4 said Melbourne, and gave its reasoning: a hook turn is a way of turning right from the left-hand lane, used at a handful of intersections in the middle of that one city and almost nowhere else on earth. Nothing in the sentence is a secret. It is a complaint about traffic.

Melbourne is incidental, and so is the hook turn. Everywhere on earth leaks something. The word for the strip of grass between the footpath and the road. Whether the price on the shelf is the price at the till. How long the summer holiday runs, which floor is the ground floor, the fact that somebody mentioned snow in July. None of it is information a person would think to keep back, and each piece of it narrows the planet a little. Write anything at all for a few years and you have handed over a few hundred of them without once saying where you live. Something reading all of them together does not need any single one to be a secret.

Behind that one line sits a real dataset: 5,814 comments posted between 2012 and early 2016, taken from 520 public Reddit accounts and handed to a model to see what could be read off them. The answer was location, income, sex, age, education, occupation, place of birth and relationship status, right on the first guess up to 85% of the time, about a hundred times cheaper than paying people to do the same reading and two hundred and forty times faster. Those comments were a decade old before anything read them. It worked on the parts nobody thinks to hide.

That line is one of three in the same example. The other two are a grumble about being overcharged, in 34d, and a memory of watching Twin Peaks after coming home from school. The model shows its working for each -- 34d is a bra size, so the writer is a woman; Twin Peaks was on in 1990 and 1991, so she was at school then -- and hands back a profile. Melbourne, female, forty-five to fifty. The prompt that does all this opens by proposing a guessing game.

Somebody who was in high school while Twin Peaks was on television was a teenager in the summer of Castlemorton.

People whose job is privacy have a word for this kind of test. Katharine Jarmul, who works on privacy in machine learning, calls it an evaluation, and files this reading of ordinary sentences alongside the older worry that a model might repeat something it memorized. The word is worth holding onto for what it quietly admits. An evaluation is an exam with a score. The machine sat it. The score was about you.

Knowing a city is still some way from knowing a name, and in 2026 a team closed that distance. They took 338 real Hacker News users who had linked their true profiles, stripped every name, handle and address out of their posts, and set agents loose to find them again among eighty-nine thousand strangers. Two-thirds came back correctly identified. The whole run cost about two thousand dollars, a dollar or two a head.

Nobody in that experiment had an enemy. No one had spent six years wondering about them, or sold a story about their clothes, or stood in a corridor after hours trying a handle. They had written about work and weather and the traffic, and something read them for less than the price of a sandwich.

In early 2021, with the clubs shut for most of a year, a disc jockey named Marea Stamper left a voicemail for a producer in London about what had been taken away. He kept the recording and built a record around her voice. The line everybody took from it was the plainest thing she said: we have lost dancing.

The clubs came back. That loss had a name, a date, an end, and a song written while it was still happening. The other thing that went in those same years had none of that. There was no night it stopped and no morning it resumed, and the people it happened to were, by the nature of it, in no position to notice.

The natural reflex here is to post less, and it arrives about a decade late. The same team tried scrubbing the texts first, running them through a commercial tool set to strip out every location, address, name, age, date and sum of money it could find. The model still pulled location and age out of what was left, off turns of phrase the scrubber had no reason to touch. Your second account was only ever unread. Your face is in the index because a friend tagged a wedding photo in 2011. Your cousin's spit is on a genealogy site. And the newest door has no lock on your side of it: when a friend plugs an assistant into their mail and messages, everything you ever sent that friend becomes something the assistant knows, and nobody asked you. You can be careful. You cannot be careful on behalf of everyone who has ever known you.

Then there are the cameras on the way home. A suburb north of Atlanta put license-plate readers on its roads and found it was sharing every passing car with more than two thousand organizations, a graph its own administrators had not known the size of. The residents never signed anything. A contract had.

Face, prose, blood, plates. Each alone is a lead. The first customer able to run all four at once is the one that already owns the warrants and the crime lab, and it does not have to want you either. It only has to be asked.

Thirty thousand people stood in a field for a week and the country could not name one of them. It spent four million pounds establishing that. The same field today would have a phone in every pocket, a plate on every car in the lane, and in every frame a face that needs nobody to recognise it. None of that would take a law, or four million pounds, or one person who wanted to know. Nobody would have to look. It would already have been read.