Sponsored by
The verification problem has changed sides this week. For most of the past year the worry has been lawyers relying on their own unchecked AI output, but the sharper story now is the AI-generated material arriving from the other party, often a litigant in person, and the time it takes to sort the real from the invented. The lead looks at what the Law Society Gazette calls law's "quiet epidemic" of AI slop, and what a represented party is supposed to do about it.
AI in Practice
AI "slop", and what to do when it lands on your desk from the other side
The Law Society Gazette this week ran an in-depth piece on what it calls a quiet epidemic: litigants in person using generative AI to produce claims and submissions, and the growing burden that places on the courts and on the lawyers who face them.
The examples the Gazette collects will be familiar to anyone in contentious practice; we have all seen the perfectly drafted letter that, whilst looking great, contains legal nonsense. One litigant produced a list of citations from ChatGPT that were, in the words of a solicitor quoted, all rubbish, and a good deal of time went into wading through them. Another filed a beautifully fluent, flawlessly formatted position statement running to seventeen pages, in which the cited case law was complete fiction.
The point that interests this writer is where the cost falls. When a represented opponent files fabricated authorities, the professional consequences at least attach to a regulated person, as the Cork v Smith admonishment and the various SRA referrals over the past year have shown. When the same material comes from a litigant in person, there is no such backstop, and the work of identifying that a case does not exist, or that a statute says nothing of the kind, lands on the other side and on the judge. Fluent formatting makes this worse rather than better: a document that looks polished inevitably invites less suspicion, so the checking is easy to skip precisely when it should matter most.
None of this is a reason to treat unrepresented parties as adversaries. The practical challenge, as the Gazette frames it, is to respond robustly to deficient AI-generated material while still treating the litigant fairly and proportionately, and keeping correspondence clear and accessible. That is a difficult balance to strike under time pressure, and it is one the profession has not really had to strike before at this volume. Sir Geoffrey Vos has described the courts as facing an "AI revolution" driven by wider access to these tools, and this is one of its less welcome faces.
For firms, the honest takeaway is that citation-checking is no longer only a discipline you apply to your own work. It is now a routine step in reading the other side's, whoever produced it.
See this week's practice prompt (at the end of this newsletter) for a suggested prompt to help with citation checking.
Takeaways
Act: When an opponent, and especially a litigant in person, files authorities or a statutory argument, verify the key cases and provisions against the primary source before you respond, and do not let polished formatting stand in for accuracy.
Watch: Whether the Civil Justice Council's work on AI in court documents, and any further steer from the judiciary, produces a practical route for flagging suspected AI-fabricated material to the court.
Risk: Responding to a litigant in person's AI-generated case carries its own conduct dimension. The reply needs to be firm on the law but fair and proportionate in tone, particularly where the person plainly does not understand that the tool has invented their best points.
Read: Law Society Gazette
On your radar
A new AI-native law firm launches, this time in the Nordics: Antti Innanen, who built the experimental agentic "firm" Lavern earlier this year, has launched Brahe, an AI-native law firm based in Helsinki and aimed at corporate and commercial work for mid-size clients. Its pitch is blunt: "Brahe runs on Claude from day one. Drafting, research, comparison, citation. Everything happens on top of AI," with human lawyers positioned as the point of the firm and AI as the operating system underneath. Why it matters for UK lawyers: this is the same structural idea as Garfield here in England and Wales, a firm designed around AI rather than one bolting it on, and it is now appearing across jurisdictions. Watch how these firms price and staff work, because that is where the competitive pressure on conventional models will show up first. (Artificial Lawyer)
Norm Ai becomes a legal-AI "unicorn" on a compliance pitch: Norm Ai, which embeds regulatory reasoning into AI agents and runs an affiliated law firm to deliver services through them, has raised a $120 million Series C at a $1.2 billion valuation led by Khosla Ventures, with a former chairman of Kirkland & Ellis among the investors. The company says the money will expand its practice-area coverage and its "supervisory agents" for regulated enterprise AI deployments. Why it matters for UK lawyers: the money is chasing agentic compliance work, and the interesting part for practitioners is the supervisory-agent framing, the idea that one AI checks another's output. Treat that as a claim to test, not a solution to trust: if a vendor offers an automated verification layer, ask how it is validated and what a human still has to sign off. (LawSites)
Harbor buys UK training firm iTrain, a sign of where the adoption problem really sits: The legal services group Harbor has acquired iTrain, a UK specialist in legal technology training, user adoption and change management that works with Magic Circle and mid-market firms. The rationale, citing this year's Wolters Kluwer Future Ready Lawyer survey, is that inadequate training remains one of the most consistently cited barriers to AI adoption, and that firms' technology spending has been outpacing the training needed to make it useful. Why it matters for UK lawyers: the constraint on getting value from AI is rarely the tool, it is whether fee earners actually change how they work. If your firm has bought AI licences, check what proportion of the budget went to training and adoption, and be honest about whether the tools are being used or just paid for. (Legal IT Insider)
Ad Break
In order to help cover the running costs of this newsletter, please check out the advert below. In line with my promises from the start, adverts will always be declared.
You’re gonna want to see this live
On July 16th at 1PM ET, beehiiv is unveiling the next chapter for audience-led businesses.
For years, creators and brands have been forced to stitch together bloated stacks of tools just to publish content, grow an audience, and make money online.
Newsletters in one platform. Websites in another. Podcasts somewhere else. Analytics scattered everywhere.
beehiiv thinks there’s a better way. Now, they’re ready to show it off at their Summer Release Event.
For Review
The Legal AI Advantage Won't Come From the Model Alone (Artificial Lawyer)
A piece by Sam Kidd of LawVu arguing that two tools running the same underlying model can produce very different results, because the advantage sits in the surrounding context: the firm's data, workflows and institutional knowledge, most of which is scattered across inboxes, drives and individual lawyers' heads. It extends last month's "scaffold" research into procurement terms, and it is a useful counterweight if your firm is choosing tools on model brand alone.
Read or listen: Artificial Lawyer
Advisory AI Growth Lab for legal services (GOV.UK)
The government's advisory sandbox for AI in legal services, which brings together the Legal Services Board, the SRA, the Council for Licensed Conveyancers and the ICO to work through cross-regulatory questions with innovators. Participation is not any form of regulatory approval, but for a firm or lawtech developer building an AI product, it is worth understanding before applications open later this summer.
Read or listen: GOV.UK
Practice Prompt
Try the below prompt to triage a set of authorities or a submission you have received from the other side, the exact task this week's lead story shows becoming routine. It is built for the case where you suspect some of the cited law may be AI-generated and needs checking before you respond. Ensure you fill in context and constraints and other aspects marked with {}. Remember to adhere to the Golden Rules and do not upload confidential or privileged information to public tools.
This author personally uses an MCP like this one to give AI tools the ability to check sources such as BAILII (publicly accessible sources only) to assist in the checking of citations. There is, of course, some irony in an AI checking AI, and must not be a replacement for proper verification.
You are assisting a civil litigation solicitor in England and Wales. Your task is to help
triage a document received from an opposing party (which may be a litigant in person) that
cites case law, legislation or rules, some of which may be inaccurate or AI-generated. This
is a planning aid only: the solicitor verifies everything against primary sources.
Context to apply:
- The document: {e.g., "17-page position statement" / "list of authorities" / "letter of claim"}
- Who produced it: {e.g., "litigant in person" / "represented opponent"}
- The matter: {claim type, court or track, the issues in dispute}
- What the document is arguing: {summarise the key propositions of law relied on}
Produce:
1. A verification worklist
Extract every case, statutory provision, rule and quotation the document relies on, and
for each set out exactly what the solicitor should check against the primary source
(does the authority exist, does it say what is claimed, is the citation correct, is it
still good law). Order the list by how load-bearing each authority is to the argument.
2. Red-flag indicators
Flag features that commonly signal AI-fabricated or unreliable content, for example
citations that do not resolve, courts or judges that do not fit the citation, quotations
that cannot be located, or propositions stated more confidently than any real authority
would support. Be clear that these are prompts to check, not findings.
3. Substantive response outline
Assuming verification confirms the concerns, outline a response that deals with the
argument on its merits and, where appropriate, brings the position to the court's
attention, while remaining fair and proportionate to a litigant in person who may not
understand that the tool invented their strongest points.
4. Tone check
Draft two or three sentences of correspondence that are firm on the law but measured and
accessible in tone, suitable for a litigant in person.
Constraints:
- {Add matter-specific constraints, for example the relevant pre-action protocol or the
court's own directions.}
- Apply the law of England and Wales only.
- Do not invent or confirm case law, statutes or rules. Your role is to identify what must
be checked, not to state the law. Where you are uncertain, say so.
- This is a planning aid, not legal advice, and the solicitor remains responsible for every
point put to the other side or the court.How did we do?
Hit reply and tell me what you would like covered in future issues or any feedback. We read every email!
Thanks for reading,
Serhan, UK Legal AI Brief
Disclaimer
Guidance and news only. Not legal advice. Always use AI tools safely.
Recommended Newsletters
Below are a few newsletters that I recommend, for various reasons. Check them out!
