What happened
According to Legal Futures, Mr Justice Sweeting, sitting in the High Court, allowed two successful litigants in person (LiPs) to recover the £1,488 cost of a subscription to an AI legal research tool. The couple, pensioners in their 70s who could not afford legal advice, had appealed a suspended possession order obtained by Wargrave on Thames Housing Association; the parties agreed the order could not stand and that the appellants were entitled to costs.
In the consequentials ruling, the disputed item was the subscription. Sweeting J rejected the argument that it was a payment for legal services under CPR 46.5, characterising it instead as expenditure incurred by the appellants in carrying out their own legal work. He then applied the ordinary disbursement test: was it properly a disbursement, and was it reasonably and proportionately incurred for conducting the litigation? He answered both in the affirmative, noting the subscription was used for research and identifying relevant authorities, that the appellants “did not use a general or free LLM” but a “sector specific artificial intelligence agent”, and that their submissions were “careful, reasonably focused and ultimately successful”. Crucially, he added: “In doing so I do not hold that the cost of AI-assisted support is automatically recoverable.”
What is new, and what it is not
This is not a new product launch, so comparisons with a previous version or a competitor are not the story. The novelty is procedural. As Sweeting J noted, there was no direct authority on the recoverability of AI research subscriptions by a LiP. The judgment treats AI as one more technology to be assessed under existing costs principles, in line with how the courts have handled electronic disclosure platforms and document management systems.
Two distinctions matter for anyone reading the headlines. First, the tool was a paid, sector-specific legal research agent rather than a general-purpose chatbot; the judge expressly recorded that the appellants did not use a free LLM. Second, the ruling is fact-specific. The subscription was bought for the litigation, funded by borrowing from a relative, and the case turned on contractual construction with potentially serious consequences for the appellants. The decision does not create a category of automatically recoverable AI spend, and it does not touch the separate question of whether solicitors’ own AI subscriptions are recoverable as between solicitor and client.
| Point | What Sweeting J decided |
|---|---|
| Nature of the cost | Expenditure by LiPs doing their own legal work, not payment for legal services under CPR 46.5 |
| Test applied | Ordinary disbursement test: properly a disbursement, reasonably and proportionately incurred |
| Amount allowed | £1,488, as claimed |
| Precedent value | None automatic; depends on the circumstances of each case |
Three concrete uses this unlocks
1. For LiPs building a costs schedule. A tenant facing a possession claim can now budget for a legal research subscription and record, from day one, why it was needed, what it was used for and how it contributed to the case. Example: keeping a short research log — questions asked, authorities found, how each fed into the statement of case — so that, if costs are awarded, the item can be justified in the same way the appellants’ was.
2. For small firms and advice agencies supporting LiPs. Clinics that cannot offer full representation can point clients towards a sector-specific research tool rather than a general chatbot, and can help them document proportionate use. Example: a housing advice service producing a one-page protocol on which tool to use, what it may be used for, and how to verify authorities before they reach a court bundle — closely tied to the discipline described in AI Hallucinated Citations: Verify Legal Research Before Court.
3. For legal AI vendors and their customers. The judgment gives vendors a defensible argument that a sector-specific agent is a litigation support tool, not a consumer gadget. Example: a barristers’ chambers or in-house team reviewing its AI stack can use this reasoning internally when justifying spend and when explaining to clients why a specialised tool was chosen over a free model. The broader adoption logic is set out in AI for lawyers.
Who and what it affects
The most immediate effect is on litigants in person and on the costs judges who assess their bills. Sweeting J’s equality-of-arms reasoning — that it “would be an odd outcome if legal professionals could deploy their standing artificial intelligence resources in support of litigation but litigants in person were, in effect, precluded from using similar tools” — gives LiPs a principled answer to the objection that AI spend is mere overhead. That objection was pressed by the housing association and rejected.
The second effect is on the economics of access to justice. A subscription that costs less than a few hours of legal advice can now, in the right case, be recovered. For advice organisations and pro bono schemes, that changes the calculus on recommending paid research tools rather than free ones. For vendors, it strengthens the case for transparent, itemised pricing that a costs judge can assess.
The third effect is cautionary. Defendants facing LiPs may now scrutinise claimed AI subscriptions more closely, testing whether the tool was genuinely used for the litigation and whether the amount was proportionate. The ruling rewards documented, responsible use and punishes the assumption that any AI receipt is recoverable.
What to watch: confidentiality, accuracy, IP and the limits of professional use
Confidentiality. The judgment does not address data protection. LiPs and firms still need to know where research queries and case facts are stored, and whether a tool trains on client material. Recoverability of a subscription says nothing about whether using it was safe.
Accuracy. The judge’s confidence rested partly on the quality of the appellants’ submissions. A subscription that produces plausible but wrong authorities would not meet the reasonableness test, and the responsible-use language in the judgment cuts both ways. Verification remains the practitioner’s job.
Intellectual property. Nothing in the ruling addresses ownership of AI-assisted work product, licensing of the underlying content, or how a tool’s terms interact with a court bundle. These remain separate risks that a costs order does not resolve.
Limits of professional use. The decision concerns LiPs doing their own legal work, not solicitors billing AI subscriptions to clients, and it creates no safe harbour. Sweeting J’s closing line is the one to quote in any internal policy: “The technology may be new, but the principles governing recoverability are not.”
How and when to act
There is nothing to sign up for. The practical step is procedural: if you are a LiP, or advising one, decide before proceedings whether a paid, sector-specific research tool is genuinely needed, record the decision and its cost, and keep evidence of how the tool contributed to the presentation of the case. If you are a firm, review how you describe AI spend internally and to clients so that any future assessment can be answered on the ordinary disbursement test rather than on novelty. And if you are building a wider AI programme, the same discipline of documented, proportionate use should sit inside it — the kind of staged approach outlined in AI Adoption in Law Firms: A Gulf First Plan (90 Days).
Sources
Frequently asked questions
Does this mean litigants in person can now always claim back AI subscription costs?
No. Mr Justice Sweeting expressly said he did not hold that the cost of AI-assisted support is automatically recoverable. Recoverability depends on the circumstances of the individual case, including the nature of the service, the purpose for which it was incurred, whether it was reasonable and proportionate, and its proper characterisation under established costs principles.
Why was the £1,488 subscription treated as a disbursement rather than legal services?
The judge rejected the argument that it was a payment for legal services within CPR 46.5. He characterised it as expenditure the appellants incurred in carrying out their own legal work, which then had to pass the ordinary test for a recoverable disbursement: properly a disbursement, and reasonably and proportionately incurred for conducting the litigation.
Does the ruling apply to general chatbots like free LLMs?
The judgment distinguishes the case before it. Sweeting J recorded that the appellants “did not use a general or free LLM” but purchased a subscription giving access to a sector-specific artificial intelligence agent used for research and identifying relevant authorities. Whether a general tool would meet the same test was not decided and would depend on the facts.
Want to apply this to your own work or your team? Tell me where you stand, and I will reply personally.
Get in touch
