AI declarations in court documents became compulsory in the Irish High Court today. Mr Justice David Barniville, President of the High Court, signed practice direction HC 142 on 29 July 2026, and it came into operation on 1 September. It reaches proceedings issued long before that date, because the test is when the document was prepared, not when the case began.
What the rule asks now it bites
The practice direction separates two kinds of document. For pleadings and submissions, being the work a legal representative puts their name to, using a generative tool does not by itself have to be disclosed. Verification carries the weight instead. Anyone relying on an authority has to confirm that it exists and supports the proposition it is cited for, and quotations have to be checked against authoritative sources and correctly attributed. Where the court directs, a party has to say whether these tools were used and how far, and everybody is expected to be able to answer that question.
Evidence is treated apart. A witness statement or affidavit has to carry a declaration in its opening paragraphs that generative AI was not used to produce the substantive content, including by "altering, embellishing, strengthening, diluting or rephrasing the evidence set out". It also confirms that the document reflects the witness's own knowledge and recollection. Formatting and spell checking are carved out of that. An expert signs a similar declaration and, where a tool has played a part in drafting, analysing or interpreting, has to set out the nature and extent of that use so the court can weigh the opinion properly.
The sanctions are set out plainly. The court may reject a document in whole or in part, make an adverse costs order, or strike out a claim or defence. A practitioner who cites a source that does not exist faces referral to the Legal Services Regulatory Authority and a personal costs order. A party who spots a problem in the other side's document raises it with them first, and takes it to the court if nothing comes back.
The checking point that catches firms out
HC 142 says that running an output through the same or another AI tool does not amount to checking it. What it allows is verification against original source material held inside the tool, which is the line between a research product that shows you the authority and a chatbot that describes one. If your people cannot open the report and read the passage, they have not verified anything.
The direction reaches past citations. Exhibits, images and recordings have to be authentic and as represented, and that duty sits on whoever is responsible for the document. A firm that takes a client's photograph or recording without asking where it came from carries the risk of it.
Documents you hold under an undertaking
Paragraph 32 catches the firm that thinks it has settled the AI question. Where documents are subject to an order of the court or to an express or implied undertaking restricting their use, generative tools must not be used on them where that would risk disclosure contrary to that restriction. Disclosure in an English action carries the same implied restriction. If your review tool sends the other side's disclosure to a service that retains what you send, the undertaking is at risk before anybody has drafted a word.
The answer is not to ban the tool. It is to know in writing what your supplier does with the material you put in, and to keep documents held under an undertaking where that answer is nothing.
What to do on this side of the Irish Sea
England and Wales has been circling the same question. The Civil Justice Council consulted on the use of AI in preparing court documents earlier this year, and its interim report proposed a declaration in language close to the wording Ireland has now adopted. Its update on the responses left professional drafting alone and marked witness statements as the area needing more work, which we set out in AI in court documents, the CJC's emerging line. A firm that reads HC 142 today has a fair sight of what a rule here would ask of it.
The work sits in how you take evidence rather than in your choice of software. Look at your statement taking process and decide where a tool is allowed near it, which for most firms means transcription and nothing beyond. Tell witnesses, in words they will follow, that a chatbot must not draft or improve their account, and keep a note that you have told them. Update your expert instruction letters so an expert knows to declare AI use and to record what was done. None of this waits on a rule change here, because a statement smoothed by a machine is a weakness in your case whether or not a practice direction says so.
The full text is published by the Courts Service of Ireland as practice direction HC 142 on the responsible use of generative AI in court documents, signed on 29 July 2026 and in force from today.
If you are unsure where AI sits in your statement taking today, a short review of that process is the sort of work we do, and it costs far less than unpicking a challenged statement after the event: talk it through with us.
