What are the consequences of misuse of AI in court cases?

Since February 2, 2025, it is a statutory obligation to promote AI literacy. The rise of AI is accelerating. Technological developments are occurring at a rapid pace. The impact of AI is also becoming increasingly noticeable in legal practice. Recent court cases show that far from everyone has a grasp on how to responsibly handle AI in legal contexts, whether it concerns the litigant, the lawyer, or even the judge.

In the article below, we provide a brief overview of recent high-profile court cases about the misuse of AI in legal proceedings. We also briefly consider the consequences.

1.    AI generated legal documents in court

A problem increasingly noticeable in the legal context is the tendency of generative AI to "fabricate" court decisions (also known as: case law) in a way that seems convincing at first glance. For example, AI-generated fake court decisions are provided with a fake identification number, typically a European Case Law Identifier ("ECLI") in Europe.  Or as another example, AI-generated fake court decisions use the name of a judge who actually exists as the person who delivered the verdict.

Only after a human check of the relevant case law registers does one discover that the ECLI number doesn't exist, or that it does exist, but the case with that ECLI number is actually a completely different case than the AI-generated text describes. Named judges, advocates - general, or legal scholars, for example, could exist but not practice at all at the court where the fake judgment was supposedly rendered, or in fact operate in a completely different area of law than how they are cited in the AI-generated fake court decision.

2.    Incorrect use of AI by the litigant

In the cases below, the judge does not appear to necessarily blame the parties to the proceedings for drafting a procedural document using generative AI, but rather for not subsequently submitting the AI-generated text to a legal professional for review.

In the case of the East Brabant District Court, dated 12 February 2026, ECLI:NL:RBOBR:2026:934, the court noted in paragraph 4.1 that: " it has the impression that the claimant, in conducting these proceedings, sought 'legal advice' from ChatGPT or another generative AI. The court finds strong evidence of this in the judgment of the Amsterdam District Court of 18 August 2022, ECLI:NL:RBAMS:2022:4703, cited by the claimant (in his reply). This judgment does not exist. [..]"

ChatGPT or another generative AI used by the claimant appears to be unaware of the facts of the case or how the relevant doctrine of Dutch law is applied.

"[.. ]Furthermore, the point the plaintiff is emphasizing in these proceedings is that, according to him, the additional assessment notice and the objection decision contain an incorrect street name, and that for that reason alone the additional assessment notice should be overturned. But the law isn't quite that formalistic, especially—as becomes clear from the further considerations—because the plaintiff knows exactly where he parked and there's no significant disagreement with the tax collector on the matter."

In the case of the East Brabant District Court, dated 24 December 2025, ECLI:NL:RBOBR:2025:8495, the judge notes in paragraphs 4.12 through 4.18, in short, that the plaintiff's argument that a previous court judgment contained a 'manifest legal or factual error' regarding the statutory requirements for hire – purchase (‘huurkoop’) was: “Contrary to what [ plaintiffs] apparently believe, neither the operative part nor the assessment in the judgment of May 1, 2025, indicates that the parties must execute a notarial deed of hire-purchase before they can proceed with the delivery of the apartment. It is important to distinguish between the notarial deed required for the delivery of the apartment and the notarial deed in which the hire-purchase agreement(s) must be recorded. Moreover, while the law requires the parties to execute a notarial deed of hire-purchase, the absence of one does not invalidate the delivery of the apartment under the hire-purchase agreement .

In short, in that case too, the generative AI tool has no firm grasp of the facts of the case, nor of the correct application of Dutch law. The AI-generated legal documents are apparently full of legal inaccuracies and irrelevant arguments that were submitted to a court without prior review.

Submitting a lawsuit full of AI hallucinations has expensive consequences

In recent rulings, courts have chosen to sanction a party that submits a procedural document generated by, or presumably with the aid of, AI to the court, for example with higher legal costs.

“The claimant would have saved the court fees—in addition to the costs already paid for the additional assessment—if he had asked someone with expertise in the matter whether it would be worthwhile to litigate this matter. It is not.” - East Brabant District Court, February 12, 2026, ECLI:NL:RBOBR:2026:934, paragraph 4.1.

Substandard AI-produced procedural documents without checking the content for legal relevance and accuracy

The consequences of the sanctions against the litigant for abuse of procedural law are particularly severe. The finding that there was abuse of procedural law in connection with the "Use of substandard AI-produced procedural documents without (having them) reviewed for legal relevance and accuracy" by the plaintiff (ECLI:NL:RBOBR:2025:8495) results in the plaintiff being ordered to reimburse the defendants' full legal costs.

“The court holds it against [the plaintiffs] that he did not properly proofread the AI-generated documents and did not have their legal relevance and correctness checked by, for example, a lawyer. It is plausible that this cost [the defendants] additional time and effort to understand [the plaintiffs]'s arguments and defend themselves against them. All things considered, the unqualified use of substandard, AI-generated procedural documents also contributes to abuse of procedural law.” – East Brabant District Court, 24 December 2025, ECLI:NL:RBOBR:2025:8495, paragraph 4.30.

Having to pay the “full” legal costs of the other party due to abuse of procedural law usually amounts to a considerably higher amount than a 'normal' award of legal costs by Dutch courts.

3.    Incorrect use of AI by the lawyer

In the aforementioned cases, the judge points out that a human legal expert could have identified the non-existence of certain rulings and the irrelevant nature of the cited doctrines to the facts of a particular case. Unfortunately, in practice, not every lawyer is so careful to review AI-generated texts before submitting them to a case.

New York: Roberto Mata vs. Avianca

One of the first cases in the world about misuse of AI by a lawyer is a case about a lawyer from New York from May 2023. The client of this American lawyer sued the airline Avianca Airline for bodily injury. The client was allegedly hit in the knee by a cart in the aisle. The problem in this case, however, was that the deadline for filing a claim had already passed. The American lawyer asked ChatGPT what exceptions there were to that deadline, and ChatGPT then fabricated non-existent cases like: "Varghese v. China Southern Airlines Co Ltd. " When asked if this was a real case, ChatGPT firmly replied that it was a real case that can be found in legal databases, in the Anglo-American context: LexisNexis and Westlaw.

Lawyers for the airline subsequently complained to the judge overseeing the case that they could not find the cited court judgments. The judge summoned the US attorney to account for six untraceable court cases and citations. The New York attorney stated that he had not known that ChatGPT could provide inaccurate information and added that he regretted using ChatGPT for legal research. To no avail, the federal judge in New York imposed sanctions on every lawyer on his team and his firm, including a fine and an order to personally inform every judge named in the fake judgments. The US judge added that he would have imposed less severe sanctions on the US attorney and his team if they had not initially tried to maintain that the fake judgments were genuine cases. The US judge added that it was up to the lawyers in question to apologize.

The Netherlands

In the Netherlands, too, lawyers recently failed to check AI-generated text for accuracy before submitting it to the court. According to the Dutch Bar Association, judges in Arnhem, Rotterdam, and Groningen raised the alarm about documents allegedly submitted by lawyers using AI.

For example, East Brabant District Court, February 12, 2026, ECLI:NL:RBOBR:2026:909, paragraph 4.5, in which the judge stated: “The court has the strong impression that the attorney for [defendant 1] had (parts of) the statement of defense drafted by an AI tool without, at least, checking the statements referred to by the AI tool to verify whether an incorrect or (as in this case) fabricated source was used. If this is indeed the case, the subdistrict court judge considers this reprehensible. This harms the client's interests and may mislead the other party to the proceedings. As discussed below, the case will be referred for a formal statement. The attorney for [defendant 1] must provide further explanation of the above in this statement.” 

In this case, the subdistrict court judge demanded that the lawyer provide an explanation, opting for a lighter sanction than that demanded by the opposing lawyer, who had requested to completely ignore the AI-generated parts of the procedural document.

Mandatory AI course

The consequence for two Dutch lawyers already is that they acted in a manner worthy of disciplinary sanctions by the board of discipline of the Dutch bar association. Two lawyers have already received official warnings on their record from the disciplinary court and are required to take a mandatory AI course.

4.    Incorrect use of AI by the judge

It's understandable that judges around the world don't appreciate substandard AI-produced documents full of legally inaccurate and irrelevant information. However, even among judges themselves, AI literacy isn't yet a given.

In the case of Gelderland District Court, dated June 7, 2024, ECLI:NL:RBGEL:2024:3636, paragraph 1 , a subdistrict court judge used ChatGPT to calculate damages, as if it were a calculator or an independent expert: "The subdistrict court judge, partly using ChatGPT, estimates the average lifespan of solar panels from 2009 at 25 to 30 years; that lifespan is therefore set here at 27.5 years. Those panels therefore still had approximately 15 years left in 2022. Over that remaining time, this therefore amounts to a maximum (and likely decreasing) loss of efficiency of 7.5 MWh . Assuming, again partly using ChatGPT, a current average kWh price of €0.30 (€0.29 to €0.34, depending on the type of contract, but [the claimant in the main proceedings] has not stated anything about this)."

The consequences for this judge were primarily widespread national criticism, including in legal journals. Anna van Duin, associate professor of private law, commented on the ruling: "Apparently, judges need to take a course.”

The lawyer for the solar panel owner noted that, according to procedural rules, a party to the proceedings must have the opportunity to comment on the investigation. His client had not been given the opportunity to respond to ChatGPT. Whether an appeal has been filed has not been publicly announced.

In a more recent case, a judge also erred. In the case of The Hague District Court, 10 October 2025, ECLI:NLRBDHA:2025:18730, paragraph 6.1, "It was explained at the hearing that, as is apparent from ChatGPT, claimant 2 in any event does not meet the material requirements for Armenian citizenship. ", instead of basing the judgment on Armenian law, a statement from an authorized representative of the Republic of Armenia, or a report from an independent expert in Armenian law, even though these sources are available.

AI output often not suitable as convincing evidence

The examples of judges incorrectly using AI as evidence to support a decision by the court are particularly unfortunate in light of other developments in case law regarding AI use. There are many recent court rulings in which judges specifically criticize parties for using AI analyses as a source of evidence to support a claim in a way that is inappropriate within the system of Dutch evidence law and Dutch procedural law, including:

Council of State

•    Council of State, 29 January 2025, ECLI:NL:RVS:2025:335, paragraph 9.2: (“At the hearing, [company name] explained that it obtained this information via ChatGPT. [company name] did not submit the prompt question. Furthermore, the information generated via ChatGPT did not provide a source reference for the answer to the question asked. The information also states that for an accurate analysis, it is advisable to contact a valuation expert with knowledge of business parks in North Limburg.” )

Court of Appeal

•    The Hague Court of Appeal, 5 March 2024, ECLI:NL:GHDHA:2024:711, paragraph 5.5;
•    The Hague Court of Appeal, 26 June 2024, ECLI:NL:GHDHA:2024:1771, paragraph 5.1.4;
•    Amsterdam Court of Appeal, 7 October 2025, ECLI:NL:GHAMS:2025:2647, paragraph 4.29: (“Finally, the court does not attach any significance to the results of consultation of ChatGPT, because it is not possible to verify the source of the information provided by ChatGPT. Moreover, the prompt question is not focused on the specific case of [name 2][..]”).

Court

•    Overijssel District Court, 28 May 2025, ECLI:NL:RBOVE:2025:3534, paragraph 3.9;
•    Amsterdam District Court, 11 June 2025 ECLI:NL:RBAMS:2025:3935, paragraph 4.11;
•    Midden – Netherlands District Court, 12 June 2025, ECLI:NLRBMNE:2025:3133, paragraph 4.9.3: (“ The court also does not agree with the claimant’s argument about the conflicting needs that give rise to her complaints, because the claimant only substantiates this with general information from websites and a text from ChatGPT”) ;
•    The Hague District Court, 20 June 2025, ECLI:NL:RBDHA:2025:10918, paragraph 10;
•    Rotterdam District Court, 9 July 2025, ECLI:NL:RBROT:2025:9914, paragraph 5.24;
•    Zeeland-West-Brabant District Court, 23 September 2025, ECLI:NL:RBZWB:2025:6499, paragraph 9.3: (“It is up to the claimant to substantiate his argument that the measured speed difference could have been caused by the circumstances under which the measurement took place. The court finds that he has failed to do so. The answer from the AI program submitted by the claimant is insufficient in this case to make the correctness of his argument plausible. It is important in this regard that the claimant has not submitted the entered prompt question, the calculation made or the source reference.”);
•    Rotterdam District Court, 23 September 2025, ECLI:NL:RBROT:2025:11087, paragraph 7: (“The information apparently generated by AI cannot be followed, as it is not known on which validated information source it is based.”);
•    The Hague District Court, 24 December 2025, ECLI:RBDHA:2025:25171, paragraph 8.
•    The Hague District Court, 9 March 2026, ECLI:NL:RBDHA:2024:18167, paragraph 10.1: (“The claimant has not substantiated this assertion. The court finds it insufficient that the claimant's representative at the hearing referred to an answer from ChatGPT as substantiation.”).

Disciplinary Board (disciplinary court)

•    Disciplinary Board Amsterdam 11 November 2025 ECLI:NL:TADRAMS:2024:191, paragraph 4.5;
•    Disciplinary Board Amsterdam, 24 November 2025, ECLI:NL:TADRAMS:2025:218, paragraph 4.3.

5.    Conclusion

In conclusion, the use of AI in legal proceedings generally doesn't work out well, even though it's not prohibited. No matter how clever the technology is and how quickly it improves, legal proceedings demand precision and accuracy from the parties to the proceedings, the lawyers, and the judge. Text with a legal function must contain legally correct, substantively accurate, and relevant information to the actual case at hand, all in accordance with the statutory rules of procedural law and the law of evidence.

For the time being, the critical eye of human legal experts remains very much needed, also to thoroughly check the homework of the AI systems before this technology can be used transparently and successfully in a legal case.

6.    Consequences of misuse of AI in your company or organization

We previously wrote about guideposts for drawing up the AI policy of your own company or organization:

https://www.wisemen.nl/nl/artikelen/deel-ii-ai-recht-hoe-maak-ik-ai-beleid-voor-mijn-eigen-bedrijf-of-organisatie-/

https://www.wisemen.nl/nl/artikelen/hoe-ontwikkelt-mijn-bedrijf-of-organisatie-ethisch-verantwoorde-en-betrouwbare-ai-/ 

The aforementioned matters relate to the impact of the rise of AI in the legal sector, but even if your company or organization operates in a different sector, it may be wise to follow these examples and consider the reasonable consequences for violating your AI policy, such as a mandatory (additional) AI course. If you were to impose a fine or more severe sanction on an employee who uses AI irresponsibly, you must be careful not to violate employment law.

7.    Questions?

The rise of AI is happening at a rapid pace. Do you need human oversight from experienced IT lawyers? Do you have questions about what the rise of AI means for you? Are you looking for guidance amidst all the developments? You don't have to face it alone.

Don't fall behind. Contact our IT lawyers today. They're happy to assist you!