The Archive Has No Statute of Limitations
 |
Forgetting used to be free. Remembering was the expensive thing.
Philosophy Newsletter / №XXVII / Monday, 21 September 2026
A man spent four years trying to make one page about him harder to find. He won, and became the most findable version of himself that has ever existed.
In January 1998, the Barcelona newspaper La Vanguardia printed a short legal notice. A property was going to auction to settle social security debts. The notice named the debtor, Mario Costeja Gonzalez. It was the kind of announcement that exists in order to be published and then ignored, a procedural artifact with a natural half-life of about a day.
Then the newspaper digitized its back catalogue. Then Google indexed it. And a decade after the auction, if you searched his name, the notice about his debt was one of the first things you found. Not a summary of his career. Not his professional listing. A repossession.
He asked the newspaper to take it down. The newspaper said no, and it was right to say no, because the notice had been published lawfully at the request of the Spanish Ministry of Labour. He asked Google to stop linking to it. Google said no. In 2010 he complained to the Spanish data protection agency, which took his side against Google and not against the paper, an asymmetry worth sitting with. Google appealed. The case climbed to the Court of Justice of the European Union, which ruled on 13 May 2014 that a search engine is a data controller, that indexing and ranking constitute processing of personal data, and that a person can in some circumstances require a search engine to delist results about them.
Now search his name. You will find the auction notice discussed in law review articles, in newspaper explainers, in European Commission fact sheets, in graduate syllabi, and in the paragraph you are reading. A 1998 debt is more legible today than at any point in its existence. He won the right to be forgotten and in the same motion became the permanent worked example of it.
I keep returning to this case, not for the irony, though the irony is perfect, but because it is the cleanest available demonstration that forgetting is not something you can request. Forgetting is a property of a system. And we have spent thirty years carefully engineering that property out.
What forgetting used to cost
I spend my working life on retrieval. Indexes, embeddings, vector stores, the machinery that decides what a system can find. That vantage makes one historical fact very hard to unsee.
For almost the whole of recorded history, forgetting was free and remembering was expensive. Clay cracked. Vellum was costly enough that monks scraped it clean and wrote over the top. Paper burned, and libraries burned with impressive regularity. Microfilm needed a reader, a cabinet, a room, and a person who knew which reel to pull. A well kept newspaper morgue in 1985 contained the past, in principle, but getting at it required a human being with institutional knowledge, an afternoon, and a reason. Every one of those was a filter.
The bottleneck was never storage. It was retrieval. An unindexed archive is a functioning form of forgetting. The document exists and cannot be reached, which for every practical purpose is the same as the document not existing. Obscurity was not a policy. It was the ambient condition, maintained for free by friction.
Two things then happened, and only one of them gets discussed. Storage got cheap, falling by something like six orders of magnitude per gigabyte between 1990 and now. That gets all the attention, and it is the less interesting half. The more consequential change is that retrieval stopped requiring you to know what you were looking for. Full text search removed the need for a catalogue. Semantic search removed the need for the right keyword. You can now surface a document you cannot name, written in a language you do not read, about a person whose name you have misspelled, on the basis of a vague recollection of its subject matter. That is not an improvement in memory. It is the systematic destruction of obscurity as a defensive layer.
Which means the right to be forgotten is not a request to delete anything. It is a request to manually reintroduce friction that the technology removed. Delisting leaves the La Vanguardia page exactly where it was. It removes the path.
That distinction matters more than the debate usually allows. The European remedy is not erasure. It is a deliberately engineered gap between what exists and what is reachable. We are asking a search company to rebuild, by hand and case by case, the accidental mercy that microfilm used to provide for nothing.
The case for forgetting as a condition of personhood
Nietzsche, in the second essay of the Genealogy of Morality, treats forgetfulness not as a defect of memory but as an active faculty, a doorkeeper who holds the door shut so that something new can happen in the room. Strip it out and there is no present tense left to live in, only an accumulating account. He was writing about guilt and debt, which makes him unexpectedly literal company for a case about a repossession notice.
Our legal systems reached the same conclusion long before anyone had a search engine to worry about. Criminal records get expunged. Juvenile files get sealed. Convictions become spent under the United Kingdom’s Rehabilitation of Offenders Act 1974, after which an applicant may lawfully answer that they have none. Bankruptcy is discharged. Civil claims die on a statute of limitations. None of these mechanisms assert that the underlying event did not occur. Each asserts something much stranger and more interesting, which is that at some point the social value of a true fact falls below its social cost, and the correct institutional response is to stop making it available.
The German Federal Constitutional Court put this most sharply in 1973, in the case arising from the Lebach killings. A broadcaster planned a dramatized documentary about the crime, scheduled close to the release of one of the men convicted as an accomplice. The court blocked it. Its reasoning was that the event’s legitimate news interest had faded with time, while the released man’s interest in being able to rejoin society had not, and that a person who has served a sentence holds a constitutional interest in not being confronted with the deed indefinitely.
Read that again with a search index in mind. The court was protecting a man from a single television broadcast on a single evening. It was not imagining a condition of permanent, global, instantaneous, self-serve confrontation available to any stranger with a phone.
The general argument runs like this. Identity is not a ledger. A person at forty five is not the same causal system as the twenty two year old who did the thing, and if we believe in rehabilitation in any sense beyond the rhetorical, we are committed to the claim that a past can become less relevant to a present. A frictionless universal record makes that claim unperformable. We can keep saying the words. We cannot keep the promise, because every employer, landlord, date, and mild acquaintance now has one click access to the version of you that the institution formally agreed to stop holding against you.
The case for the duty to remember
Now the other side, which is stronger than its opponents usually grant, and which I find harder to argue against the longer I look at it.
Every apparatus that has ever done something monstrous has wanted the record gone. In the winter of 1989 and into 1990, ordinary citizens occupied Stasi offices across East Germany specifically to stop files being shredded, and the reconstruction of what they saved is still going on decades later. In July 2005, investigators looking into a munitions depot in Guatemala City found the abandoned archive of the National Police, tens of millions of pages, water damaged, rat infested, unlisted, unknown. That archive became evidence. People were convicted on the strength of paperwork that a state had assumed would rot.
Against that, “the past should become harder to find” starts to sound less like mercy and more like a service offered to whoever is best positioned to buy it.
The empirical objection is softer than the historical one. Google’s transparency reporting has consistently indicated that the large majority of delisting requests, on the order of nine in ten, come from private individuals asking about ordinary personal information rather than from public figures managing reputations. I would want anyone quoting that figure to check the current report rather than take it from me. Read at face value it supports the first position handsomely. The trouble is that a policy is not judged by its median case. It is judged by its contested ones, and the contested ones are a physician with a malpractice finding, a company director with a fraud conviction, a candidate for office with a history.
And then the structural objection, which I think is close to unanswerable. The European mechanism hands the weighing to the company. Google receives the request. Google balances public interest against private harm. Google decides. There is no hearing. In the ordinary case there is no published reasoning, no cross examination, no discoverable standard, and no appeal that a person of normal means can afford to pursue. We took a question that constitutional courts find genuinely, visibly difficult, and we routed it into a moderation queue with a response time target.
Paul Ricoeur, writing about memory and forgetting at the end of the last century, argued that both excessive remembering and excessive forgetting are pathologies, and that what a society owes its past is not maximal recall but a just memory. The trouble with justice as a criterion is that it requires a judge.
The case that breaks both positions
Let me construct one, and flag it as constructed, because the real cases I know well enough to describe accurately are all too convenient for one side.
A nurse makes a medication error. A patient dies. She is struck off, cooperates fully with the inquiry, and does not contest the finding. Twelve years pass. She has retrained into an unrelated field, has never sought reinstatement, and has done everything that any theory of rehabilitation would ask of her. Her name plus the word “inquest” is the first result about her, permanently, and it will be the first result when her children search for her.
Position one says: delist it. She has served her professional death sentence, the regulator’s register still holds the formal record for anyone with a legitimate need, and the search result serves no function except recreational judgment.
Position two says: this is precisely the information a patient safety researcher needs, precisely the pattern that only becomes visible in aggregate, and the confident assertion that nobody will ever need it is exactly what the Guatemalan police assumed about their filing cabinets.
Both are right. That is the problem. And I think the reason both are right is that each position is quietly assuming there is one public, with one legitimate interest, and a single correct setting for how findable a fact should be.
There is no single public. Helen Nissenbaum’s framing of privacy as contextual integrity is the sharpest instrument available here. Information carries norms from the context that produced it. A regulator’s register is a context, with a purpose, an audience, and an implicit set of rules about what the record is for. A search result is a different context, and moving the fact from one to the other strips the norms off it while leaving every word intact. The violation is not that the information exists, and not that it is true. The violation is that it travelled, and arrived somewhere that its original context never licensed.
Which suggests the right to be forgotten may be the wrong shape for the problem. It asks whether a fact should be findable, a binary, globally. The interesting question is who should be able to find it, for what, and whether a system can be built that carries a fact’s context along with the fact instead of discarding it at the index boundary. I do not know how to build that. I am not certain it is buildable at all, since the whole commercial value of a general index is its refusal to care where a document came from.
Where I am unsure
- The nine in ten figure for private individual delisting requests is drawn from Google’s transparency reporting as I last read it. Treat it as approximate and check the live report before repeating it.
- The Lebach line of German cases did not stay settled. A later decision in the same lineage came out the other way for a different film, which means I am presenting a snapshot of a doctrine that has moved.
- I have described the Guatemalan National Police archive from published accounts of its discovery and use, not from primary sources, and the page counts quoted in the literature vary.
- The nurse is invented. No part of that case is a real person.
- My claim that semantic retrieval matters more than cheap storage is an argument, not a measurement. I believe it. I cannot prove it with a number.
We have built the first civilization in history whose past does not decay. Not by deciding to. By making storage nearly free and retrieval nearly frictionless, and then discovering, somewhere around 2010, that we had removed a mechanism nobody had thought to name because nobody had ever had to pay for it.
The single forgetting mechanism we then built to compensate is a web form, reviewed by a company, on a schedule.
So the question is not what we owe Mario Costeja Gonzalez, who has after all won. It is what happens to the rest of us. Each of us has a 1998 notice out there, or will. And each of us will eventually face the choice he faced, which is whether to leave it sitting in the dark, or to make it famous by asking for it back.
Comments
Post a Comment