Practical examples, warnings, and analysis on generative AI use in the legal sector.
Dutch courts and disciplinary bodies have repeatedly encountered AI-generated legal texts containing incorrect or fabricated citations.
30 July 2026
The Disciplinary Board in ’s-Hertogenbosch reprimanded a lawyer who copied non-existent and incorrect case law from an AI tool into two tenancy cases. One ECLI number did not exist; seven others pointed to completely different cases. The lawyer had not checked the generated sources.
3 June 2026
BNR analysed recent case law and found non-existent jurisprudence or other AI-generated materials from citizens and lawyers in at least thirteen cases. Judges are spotting this more often, and the pressure on the courts is increasing.
11 May 2026
Delissen Martens outlines how judges assess AI output: sources must be traceable, claims verifiable, and article 21 of the Dutch Code of Civil Procedure remains leading. There is no AI excuse for incorrect or fabricated citations.
9 April 2026
At the Enterprise Chamber, a lawyer cited non-existent case law in a shareholder dispute. Judges spoke of AI that had “run amok”. Corrections produced new fictional ECLI numbers. The case ended in a settlement.
16 March 2026
A Rotterdam lawyer previously called out for AI-hallucinated case law again submitted non-existent judgments at the same court. The court disregarded the reliance on that purported case law.
1 March 2026
In an Eindhoven tenancy case, a lawyer withdrew eight ECLI citations because they could not be traced. One number did not exist; seven others belonged to completely different cases. The judge demanded an explanation of suspected AI use.
26 February 2026
The Innovative Lawyer surveys Dutch cases involving “ChatGPT jurisprudence”: from disregarded claims to abuse of process and a marriage that never became legally valid because of a ChatGPT-written ceremony speech.
24 February 2026
A cantonal judge at the Oost-Brabant court ordered a lawyer to explain by submission after incorrect ECLI numbers were withdrawn in a tenancy case. Judgments had often stopped at a reprimand; here the court went a step further.
22 February 2026
Three lawyers received a warning and two had to take a mandatory course after citing non-existent judgments generated by ChatGPT in court filings. Judges in Arnhem, Rotterdam, and Groningen spotted the errors.
16 February 2026
Many Dutch lawyers report spending time every day correcting AI-generated client texts that regularly contain errors in statute articles, deadlines, and citations.
4 February 2026
Stibbe analyses when substandard AI court filings contribute to abuse of process. AI is allowed, but creates a verification duty: every citation to case law or statute must be checked individually.
7 January 2026
Several Dutch cases show that uncritical AI use can lead to inadmissibility or extra procedural costs. Judges have described such work as legal shoddy work.
5 August 2025
AI-Forum describes how persuasively written but non-existent case law can influence proceedings unnoticed. The analysis stresses that fabricated sources are not always spotted immediately, so structural source checks remain essential.
Both the Netherlands Bar Association and the Council for the Judiciary have warned about the risks of generative AI when drafting pleadings, advice, and court filings.
Courts outside the Netherlands are also alert to AI-generated legal texts with fabricated or incorrect citations.
15 February 2024
A Canadian judge published an explicit warning to lawyers not to trust generative AI blindly when drafting court filings.
13 February 2024
A lawyer used AI for a court procedure and cited multiple fictional judgments. The court spoke of gross negligence.
27 May 2023
A lawyer submitted a filing with six non-existent cases invented by ChatGPT. The judge called it absurd and considered sanctions.
As a legal professional, you remain responsible for the content you submit or sign. VeriLeges is a fixed, reproducible verification layer that checks AI texts for incorrect citations - without using generative AI itself.
Fabricated case law is one risk. A second is subtler: a real judgment that was later quashed in the same case. Where metadata allows, VeriLeges can show that as a review signal - so you do not miss that check among dozens of citations.
Start free and see which signals you get within the current verification scope.
Start free with VeriLeges