When AI goes wrong in court: the UK cases involving fake authorities and hallucinated law

Fake law is reaching real courtrooms

Artificial intelligence is already embedded in how lawyers and litigants research, draft and prepare cases. What the courts are now dealing with is the consequence of relying on it without checking the output.

Over the past 18 months a series of UK judgments have involved fake authorities, incorrect citations and legal propositions that appear to have been generated or shaped by AI. Some have ended in referrals to professional regulators. Others have produced public judicial warnings to lawyers and litigants about putting AI-generated material before a judge.

The scale is no longer anecdotal. In August 2026 the Solicitors Regulation Authority said it had received 42 reports of potential AI-related wrongdoing between July 2025 and July 2026, covering inaccurate legal citations, supervision failures and confidentiality concerns.

Ayinde and Al-Haroun: fake cases reach the High Court

The most significant warning came from the Divisional Court in Ayinde v London Borough of Haringey and Al-Haroun v Qatar National Bank [2025] EWHC 1383 (Admin), which considered two separate sets of proceedings where false authorities had been placed before the court.

In the earlier Ayinde judgment, Mr Justice Ritchie identified several authorities relied on in written submissions that simply did not exist — supposed cases including R (Ibrahim) v London Borough of Haringey, R (H) v Ealing London Borough Council and R (KN) v Barnet LBC. The judge described one of them plainly as "a fake".

It is worth being precise about what the court did and did not find. There was no finding that the barrister involved, Sarah Forey, had used AI. AI was raised as one possible explanation for how the authorities came to appear, but the judge said he could not determine whether that was what had happened. What was established is that non-existent cases had been put in submissions to the court.

That was serious enough for the matters to be considered under the court's Hamid jurisdiction, which allows the court to address professional conduct capable of interfering with the administration of justice. The later Divisional Court judgment issued a broader warning about generative AI and legal research, and restated the basic point: lawyers remain personally responsible for verifying what they file.

Re A, B, C, D: hallucinations in the Family Court

The pattern continued into 2026. In Re A, B, C, D (Extension of assessment; Use of AI: hallucinations) [2026] EWFC 71 (B), Recorder Howard dealt head-on with the use of AI in family proceedings.

A lay advocate, Layla Parsons, had prepared a skeleton argument containing authorities and legal propositions that were not genuine. She accepted she had used an AI tool to help prepare the document. The judge considered the issue serious enough to justify publishing the judgment, and Parsons was subsequently named. The Law Society Gazette reported that the skeleton contained four cases or propositions that were not real.

The wider significance is not the individual error. It is how easily fabricated law enters proceedings when AI research is treated as a finished product rather than a first draft requiring independent verification.

Rodney v Gee'z Micro Bar: solicitors referred to the SRA

The consequences hardened in Rodney v Gee'z Micro Bar and Pitstop, heard in the County Court at Dudley in April 2026, where three documents submitted for a party contained incorrect and misleading citations.

The material raised the suspicion that AI had been used to produce the research, though the court made no definitive finding that a particular tool had generated it. One solicitor accepted that a paralegal may have used AI, and accepted responsibility for failing to check the research properly.

His Honour Judge Grimshaw concluded the failures were serious enough to refer two solicitors, Mahmood Hussain and Kossar Qureshi, to the SRA. As the Gazette reported, the judge described the submission of false authorities as an increasingly serious threat to the integrity of the justice system.

The distinction matters for firms. The problem was not that AI may have been used. It was that unreliable material reached the court without adequate supervision or verification.

Kamal v Tax Policy Associates: hallucinations as litigation conduct

AI-generated references have also surfaced outside conventional research exercises. In Kamal v Tax Policy Associates Ltd and another [2026] EWHC 551 (KB), Mrs Justice Collins Rice considered an £8m defamation claim and, in examining the claimant's conduct of the litigation, referred specifically to AI-generated "hallucinatory" references to non-existent cases, which caused the defendants extra work because the authorities had to be checked.

False authorities were not the sole reason for the outcome, but they became part of the court's wider assessment of how the litigation had been run. That is the point most easily missed: AI errors do not sit in isolation. They generate cost, cause delay and colour how a party is judged.

The courts are not banning AI

None of these judgments suggests lawyers cannot use artificial intelligence. The message is far more basic: AI does not change who is responsible for what goes before a judge.

A solicitor cannot rely on the fact that research was produced by a paralegal, an AI platform or an automated research tool if inaccurate material is then filed. The same applies to barristers and to litigants in person.

Generative AI is uniquely risky here because it produces legal writing that looks entirely plausible. A fabricated citation can resemble a genuine neutral citation. A false judgment can be described in exactly the language and structure lawyers expect. Plausibility is the trap, and verification is the only answer.

Our view: firms need controls, not bans

The practical question for firms is supervision, and it comes down to four things worth auditing this quarter.

• Is AI-generated research always checked against a recognised legal database or the original judgment?
• Do junior lawyers understand that a case existing is not enough if it does not support the proposition it is cited for?
• Do supervisors know when AI has been used in preparing work?
• Is there a written policy setting out which AI tools may be used for legal research, and for what?

The biggest risk is not a lawyer deliberately inventing a case. It is a lawyer accepting something convincing and assuming somebody further up or down the chain has already checked it. The growing body of UK case law shows exactly where that assumption ends.

AI will keep getting more useful in legal practice. But when a document reaches the courtroom, responsibility for its accuracy still belongs to the person who put it there.