On June 6, 2025, the High Court of England and Wales issued a ruling tying together two cases in which lawyers cited non-existent case law, and it delivered an unusually blunt warning about generative AI in legal work. TechCrunch covered the ruling on June 7, emphasizing that lawyers could face “severe” penalties next.
For developers and product teams building AI tools, the document is worth reading in full. It is not another AI-hallucination anecdote: a senior judge describes, in plain words, what these tools can and cannot do, and lays out a sanctions ladder that runs from professional discipline all the way to police referral. RAG, citation generation, and reference verification now have a courtroom-grade test standard.
Two cases: 18 of 45 citations did not exist
The ruling, issued by the Divisional Court of the King’s Bench Division and cited as [2025] EWHC 1383 (Admin), covers Ayinde v London Borough of Haringey and Al-Haroun v Qatar National Bank. In one case, a lawyer representing a man seeking damages against two banks filed a document with 45 case citations — 18 of which did not exist. The others, the court found, “did not contain the quotations that were attributed to them, did not support the propositions for which they were cited, and did not have any relevance to the subject matter.” In the other case, a lawyer acting for a man evicted from his London home cited five cases that did not appear to exist. She denied using AI and suggested the citations may have come from AI-generated summaries seen in “Google or Safari.” In other words, the common thread in both cases was not that lawyers used AI, but that nobody checked what the AI handed over.
The judge on generative AI: plausible but wrong
Judge Victoria Sharp wrote that generative AI tools like ChatGPT “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,” and “may make confident assertions that are simply untrue.” This does not mean lawyers cannot use AI for research, she wrote — but they have a professional duty “to check the accuracy of such research by reference to authoritative sources, before using it in the course of their professional work.” She also observed that the number of such cases is growing, including in the United States, where lawyers representing major AI platforms have cited AI-generated falsehoods, and that “more needs to be done” to ensure the existing guidance is followed and lawyers comply with their duties to the court.
Sanctions: from public admonition to police referral
The court decided not to initiate contempt proceedings in these instances, but stressed this is “not a precedent”: lawyers who fail to meet their professional obligations “risk severe sanction.” The powers available range from public admonition and costs orders to contempt proceedings and even “referral to the police.” Both lawyers were either referred to professional regulators or referred themselves, and the ruling will be forwarded to bodies including the Bar Council and the Law Society, forming the basis for further guidance and disciplinary processes. The court’s choice is worth noting: it deliberately withheld the heaviest sanction this time, but forwarded the ruling publicly to professional bodies — putting the weight of the warning ahead of the punishment. Hallucinations will not disappear because of one ruling, but acceptance criteria will change.
What this means for AI tool builders
First, fluency is no longer a quality bar. The judge’s phrasing — coherent, plausible, potentially entirely wrong — is one of the sharpest official descriptions of hallucination to date. Teams building citation, quotation, or literature-search features should treat “every reference traces back to an authoritative source” as an acceptance criterion, not a nice-to-have. Second, verification duty sits with the user, but design responsibility sits with the vendor: the ruling assigns the checking obligation to professionals, and other professional regulators are likely to follow with similar guidance. Tools with built-in source verification, citation previews, and graceful refusal will carry real compliance value over generate-only tools. Third, the risk level is rising: from the 2023 fines on US lawyers over ChatGPT-fabricated cases to a 2025 UK court naming the problem and flagging possible police referral, AI citation reliability has moved from an ethics topic to a legal-risk topic. In practice, that means citation chains in retrieval-augmented generation, visible source snippets, and line-by-line verification features are turning from research projects into product requirements, and legal and compliance teams buying AI tools will quickly move citation checking from a bonus to a requirement.
Sources
- Ayinde -v- London Borough of Haringey, and Al-Haroun -v- Qatar National Bank — Courts and Tribunals Judiciary
- Lawyers could face ‘severe’ penalties for fake AI-generated citations, UK court warns — TechCrunch
AI-assisted summary compiled from the sources above, reviewed by a human before publishing.