# AI in the family courts: the cases you haven't heard about yet…
Welcome to my Substack. Over time I'll be exploring my family law practice through the lens of the fast developing, and frankly fascinating, world of AI. Some pieces will be practical, some will be case commentary, and some will just be me thinking out loud about where this all goes next. This first one is a case commentary, and it seemed the right place to start.
Everyone's talking about the Home Office asylum case this week. A senior judge found that the document the Home Office relied on to refuse a woman's asylum claim, a Country Policy Information Note said to confirm Morocco was safe for her return, appears never to have existed at all. The refusal letter, the judge said, bears the hallmarks of AI use. It's a striking story, and rightly getting attention.
But while that's been doing the rounds, something quieter has been happening a few floors down, in the family courts. Over the past year or so, a small but telling body of case law has built up around AI, and it doesn't fit neatly into either "ban it" or "embrace it." It's more interesting than that. So I thought it worth setting out properly, for anyone practising in this area or simply keeping an eye on where things are heading.
## When AI earned its keep
I will start with the good news, because there is some. In *Re B (Fact Finding: Use of AI for Judgment Summary for Parents with Learning Difficulties)* [2026] EWFC 107 (B) ([discussed by the Transparency Project here](https://transparencyproject.org.uk/make-it-simple-make-it-clear-using-ai-to-improve-clarity-and-understanding-of-a-judgment/)), Her Honour Judge Hesford took the unusual step of using a secure judicial AI tool, built on Microsoft Copilot, to produce plain language, bullet point summaries of a fact finding judgment. This was a care case, arising from a skull fracture suffered by an eight month old child while in the parents' care, and the parents in question had learning difficulties that made the standard judgment format hard for them to follow.
Every advocate in the case agreed the summaries were immensely useful. It's one of the first recorded instances of a judge openly stating that a UK court ruling was produced by them in part with generative AI, and it's worth noting precisely why it worked: the tool was secure and judicially controlled, it was used for translation and accessibility rather than for reaching or justifying the decision itself, and the substantive judgment was the judge's own reasoning throughout. AI assisted communication of a decision that had already been made by a human being who took full responsibility for it.
That's the model worth holding onto. Not AI instead of the judge, but AI helping the judge's decision land properly with the people it affects and who needed a simplified version of it .
## When the line got tested
Then there's *D (a child) (recusal)*, which went to the Court of Appeal ([reported in the Law Gazette here](https://www.lawgazette.co.uk/news/mothers-use-of-ai-entirely-understandable-court-of-appeal-finds/5125382.article)). A litigant in person, a mother without legal representation, had relied on AI generated legal argument in presenting her case. Lord Justice Baker, with whom Cobb LJ ( as he then was; now President of the Family Division ) and Miles LJJ agreed, found this "entirely understandable" given her position, and said as much in terms: litigants in person are in a difficult enough spot when it comes to putting forward legal argument, and it's no great surprise they turn to whatever help they can find.
But he didn't stop there. The judgment was equally clear that used properly and responsibly, AI can be of assistance to litigants and lawyers preparing cases, but it is not an authoritative or infallible body of legal knowledge. And critically, all parties, represented or not, owe a duty to the court to ensure that any case cited in argument is genuine and supports the point being made. Sympathy for the litigant did not translate into any softening of that duty.
This is the case I keep coming back to when lecturing and I think about where the line properly sits. Understanding why someone reached for AI is not the same as excusing what it produced unchecked. Both things can be true at once, and the Court of Appeal managed to hold both without contradiction.
## When nobody could quite untangle it
The third case is the most unsettling of the three, in some ways, precisely because nothing dramatic happened in it. *M v F (Fact Finding Hearing)* [2026] EWFC 22 (B) ([discussed in Matthew Lee's commentary for Doughty Street here](https://insights.doughtystreet.co.uk/post/102mi96/38-uk-cases-involving-hallucinations-ai-or-otherwise-judicial-caution-in-the-f)) is reported to have involved a witness who had been assisted by AI in preparing a witness statement. According to that reporting, the witness was questioned in detail, at length, about exactly how the tool had been used, and by the end of that questioning, the judge still wasn't entirely clear on the process involved. The full judgment isn’t yet published , so I'd treat the detail with the usual caution until it's confirmed against the transcript, but the account is consistent with the wider pattern in the other two cases.
That's the bit worth sitting with. This wasn't a case of an obviously fabricated citation, easy to spot and easy to strike out. It was something more diffuse: a statement that had gone through an AI assisted process nobody, including the person who used it, could fully reconstruct or account for afterwards. The judgment reflects real, and I think entirely warranted, caution about the weight that could properly be placed on evidence prepared that way. If you can't explain how something was made, you can't properly test it, and if you can't test it, a court is right to be careful about how much it leans on it.
## Where the guidance sits
None of this is happening in a vacuum. The Courts and Tribunals Judiciary issued formal guidance on AI use in proceedings back in October 2025, and it's a measured, sensible document as these things go. It doesn't pretend AI is going away, and it doesn't need to. It permits AI for administrative and preparatory work, drafting, summarising, research support, while keeping human responsibility for the substance squarely where it belongs.
The Nuffield Family Justice Observatory has been doing valuable work in the same space, publishing an updated briefing earlier this year on AI in the family justice system. It splits current use into three sensible buckets: improving families' experience of the system, administrative efficiency, and decision support. And it makes a point I think practitioners sometimes forget in the rush to either champion or resist these tools: any of this needs to be informed by what parents and children want, not just what's technically possible or administratively convenient.
## What I take from all this
Put the asylum case and these three family court cases side by side, and a shape starts to emerge. AI earns its place doing the things it's good at: translating, summarising, drafting, searching, making decisions more accessible to the people affected by them. Where it falls down, every time, is when it's asked to stand in for the verification and judgement that a human decision maker is legally and professionally obliged to bring. A CPIN that doesn't exist. A citation that doesn't check out. A process nobody can quite reconstruct after the fact.
That obligation to verify doesn't move, whatever tool sits on the desk. It was there before AI, and it'll still be there long after today's tools look as dated as fax machines . The profession's job isn't to pick a side in some AI versus lawyers contest that was never really the right question. It's to work out, case by case, exactly where the human has to stay in charge, and to be honest when that line gets blurred, whether that's in an asylum refusal letter or a family court witness statement.
That's the conversation worth having. Not whether AI belongs in our courts, it plainly already does, but where the human in the loop has to remain non-negotiable.
NB : The full judgments are available free of charge via the National Archives' (https://caselaw.nationalarchives.gov.uk/) service, searchable by neutral citation.