AI and legal privilege: what the two judgments actually say
Two decisions now govern how a solicitor may use AI on a client's file. One is about checking what the machine tells you. The other is about where the documents go. Neither says do not use AI.
Published 14 October 2026 · Reading time 4 minutes
The first: you must check what AI gives you
In June 2025 the Divisional Court handed down R (Ayinde) v London Borough of Haringey and Al-Haroun v Qatar National Bank [2025] EWHC 1383 (Admin). Both cases involved citations to authorities that did not exist.
The court put the duty plainly:
Those who use artificial intelligence to conduct legal research notwithstanding these risks have a professional duty therefore to check the accuracy of such research by reference to authoritative sources, before using it in the course of their professional work.Ayinde v Haringey [2025] EWHC 1383 (Admin), at [7]
Three things follow from that sentence. The duty is personal, and it does not transfer to the tool. It applies whether the lawyer ran the search or inherited the output from someone else. And it is discharged by checking against an authoritative source, not by a feeling that the answer looked right.
The court also made clear that unchecked material placed before a court is a matter for wasted costs and for the regulator.
The second: an open AI tool can waive privilege
In UK and R (Munir) v Secretary of State for the Home Department [2026] UKUT 81 (IAC), an adviser had pasted client letters and Home Office decision letters into ChatGPT to summarise them. The Upper Tribunal said:
To put client letters and decision letters from the Home Office into an open source AI tool, such as ChatGPT, is to place this information on the internet in the public domain, and thus to breach client confidentiality and waive legal privilege.UK v SSHD [2026] UKUT 81 (IAC), at [21]
The same paragraph draws the line on the other side:
Closed source AI tools which do not place information in the public domain, such as Microsoft Copilot, are available for tasks such as summarising without these risks.UK v SSHD [2026] UKUT 81 (IAC), at [21]
So the question is not whether AI touched the document. It is whether the document left the firm's control. A closed, contracted tool that does not publish what it is given does not carry the same risk.
What this means in practice
- Consumer chat tools are the problem, not AI as a category. Pasting a privileged document into a public tool is the act the Tribunal described.
- What matters is the contract and the architecture: where the data sits, whether it is retained, whether it trains a model, and who else can reach it.
- Verification is not optional, and it is not delegable. A tool that makes checking quick is doing the part of the job the court cares about.
- Both points belong in a firm's AI policy, with a named person answerable for it.
Where Tristum sits
Each firm's matters run on that firm's own encrypted server in the UK, with no other firm on the machine. Connectors are read-only, so the evidence stays in the firm's systems. Only the passages needed to answer a question go to a model, on European infrastructure under contract with zero retention, and nothing is used to train anything.
Every answer is footnoted to the page it came from, and each quotation is checked against its source document before it is shown. The verification the court expects takes one click, and the audit log records that it happened.
This page describes how Tristum is built and summarises two public judgments. It is not legal advice, and it is not a substitute for your firm's own assessment of privilege in a particular matter. Both judgments are linked above and are worth reading in full.
Book a demonstration