UK Prosecutor Cited a Fake Case From an AI Summary in an 11-Year-Old's Trial── A source is checked only once it is found in an authoritative original, and responsibility for an AI-supplied citation stays with its user

On 3 November 2026, the Guardian reported that a senior prosecutor at the Crown Prosecution Service had cited a case that does not exist, in youth court proceedings over the prosecution of an 11-year-old boy. The case appears to have come from an AI summary shown at the top of an internet search. When a professional uses a case that an AI search summary supplied, what counts as having checked it? Only finding the case in an authoritative source and reading the judgment counts. England's High Court called that check a professional duty in 2025. Responsibility for an AI-supplied source stays with the person who uses it.

01The case the prosecutor cited was a number that, by the Guardian's account, existed only in an AI summary

The case cited by the senior crown prosecutor is not in any law report. The Crown Prosecution Service, or CPS, decides who is charged with crimes in England and Wales. The Guardian understands that when the prosecutor searched the internet, the search page produced an AI summary, and that summary contained the details of a fictional case.

The duty to check against the original had been spelled out well before this. In a June 2025 judgment, the High Court said that lawyers who use AI for research have a professional duty to check what it gives them against authoritative sources. The High Court is the court in England and Wales that hears serious civil cases and reviews the decisions of public bodies.

To check, in this sense, is to find the source in an authoritative one. So a person who uses a source supplied by AI has not checked it until they have found it in the original. An AI summary on a search page may have the form of a citation, but it is not the source. In my view, what went missing here was one step: going from the summary to the original.

02The citation did not exist; the court found the explanation inadequate but left the outcome unchanged

The prosecutor cited a case given as "CPS – R v J [2024] EWCA Crim 3529". EWCA Crim is the label attached to judgments of the criminal division of the Court of Appeal. According to the Guardian, no judgment with that number exists.

The sequence can be followed from the Guardian's report. In April 2026 the prosecutor decided the prosecution should go ahead, and the chief crown prosecutor confirmed that decision. The boy's lawyers challenged the decision to keep prosecuting him. In youth court arguments in that challenge, the prosecutor cited the fictional case. In October she admitted the citation had been made in error, and said she had meant a real decision from about fifteen years earlier.

Two senior High Court judges called the CPS's explanation unsatisfactory. They nonetheless found that the fictitious case did not affect the outcome of the boy's challenge to his continued prosecution. His trial is listed in the youth courts.

One distinction matters here. The point that the case came from an AI search summary is what the Guardian understands from its own reporting. Brian Flanagan, a legal scholar at Maynooth University in Ireland, told the Guardian that the legal source does not exist, and that this is a serious failure.

03Non-existent sources have appeared outside court too, including a US government health report

This was not one prosecutor's slip. In the 2025 Ayinde case, a barrister in the pupillage stage of training filed papers in the High Court that cited five cases which did not exist. The court referred the matter to the profession's regulator.

The same thing has happened outside the courtroom. A US government report on children's health, published in May 2025 and known as the MAHA report, contained 522 citations. NBC News confirmed that four of the papers it cited did not exist. The report was later corrected, and citations on advertising to children and on medication use were replaced.

Whether AI was used to write the MAHA report has not been officially confirmed. Even so, the three examples in the table share something. Someone writing a document that cites sources put in cases or papers that do not exist, without opening the originals. Court papers and government reports alike need the same check whenever they cite.

04Generative AI can produce a neatly formatted case citation with no source behind it

False sources turn up in court papers and in government documents. Why does a number that does not exist look real? The High Court's June 2025 judgment addresses exactly this.

According to the court, freely available generative AI tools are not capable of conducting reliable legal research. Such tools can produce apparently coherent and plausible responses to prompts. But those coherent and plausible responses may turn out to be entirely incorrect, the judgment said.

Case citations follow a fixed pattern. The year, the court's label and a running number always appear in the same order. A generative AI tool can reproduce that pattern, so the numbers it produces look tidy. A tidy number is still no evidence that the judgment exists.

There is only one way to check. Look the number up in a law report or a court database, and read the judgment. Only once the text is in front of you can you judge whether the case exists and whether it supports the point being made.

05The CPS ran an internal review and says case law must come from trusted legal sources

A well-formed number need not be a real one. In its statement after the error came to light, the CPS addressed where case law should come from. According to the Guardian, the CPS said the error had arisen from legal research carried out on the internet. It then said three things.

This was not the first such error at the CPS. In February 2026 the CPS acknowledged that, in an extradition case involving two Romanians, a lawyer had cited non-existent legal authorities, likely caused by AI. Legal Cheek, a legal news site, reports that the two fake cases first appeared in the CPS's grounds of opposition. They were then carried forward into a further document drafted by junior counsel.

What the two episodes share is that a case written down somewhere was copied into the next document without being checked against the original. This time it went from a search summary into court arguments; in February it went from one CPS document into another. The CPS already had a policy that AI would not make decisions and that humans would check accuracy. The fake case still reached a courtroom. Having a policy does not remove the work of pulling the original, case by case.

06Three things to separate: a well-formed citation, the duty to check, and a remembered case

The CPS policy already said where case law should come from, and the error happened anyway. From the facts so far, there are three things for anyone who uses AI-supplied sources to keep apart.

First, an AI answer can produce something in the form of a citation without guaranteeing that the source exists. As the High Court put it, generative AI answers can be plausible and still entirely wrong. How well formed a citation looks is no reason to trust it.

Second, checking against the original is already a professional duty. The 2025 judgment said lawyers who research with AI have a professional duty to check against authoritative sources. It compared this to the responsibility of a lawyer who relies on the work of a trainee. Guidance for judges issued by the judiciary of England and Wales in October 2025 also warns that AI can generate incorrect or misleading information, and stresses personal responsibility for all material produced in one's name. Using AI is no excuse for a false citation.

Third, a case you remember is not checked until you look it up again in the original. The prosecutor said she had meant a real decision from about fifteen years earlier. My reading is that she searched for a case she remembered and used the number in the summary the search page showed her. That is a guess. If it is right, opening the real judgment in a law report would have stopped the error.

07The UK tracker has reached 101 entries, and whether new rules will stop repeats is unknown

Cases where those three points were not followed are not isolated. Matthew Lee, a barrister at Doughty Street Chambers, keeps a record of AI-related errors in UK courts. According to the Guardian, the tracker now lists 101 entries. It covers confirmed, suspected and disputed errors as well as related false citations.

In a post Lee wrote in February 2026, the tally stood at forty-seven. The two figures may not have been counted in the same way, so it would be unsound to calculate a rate of growth from them.

Whether the CPS's rules will stop a repeat is not yet known. That will only become clear from future filings and judgments. My own expectation is that as long as search pages show AI summaries, the same kind of error will keep appearing outside law too. A research report citing papers, or an internal document citing rules, can go wrong in exactly the same way.

What a user can do is already clear. Look up any case or paper that AI supplies in the original source. If it cannot be found, do not cite it. If it can, read it and make sure the text supports the point you are making.

A professional who uses a case from an AI search summary has checked it only after finding it in an authoritative original and reading it. The High Court wrote that duty down in 2025. However tidy the AI's citation, responsibility remains with its user.