Does using an AI tool on client papers put privilege at risk?

Using an AI tool on privileged material is a confidentiality question before it is anything else. Privileged documents stay privileged as long as confidentiality in them is maintained, which is why firms can already send papers to counsel, experts, costs draftsmen and cloud document systems. The risk with AI is not the technology in the abstract; it is using a tool whose terms, storage and access arrangements you have not checked, or letting staff paste material into personal accounts nobody knows about. What follows is practice guidance on how to manage that, not advice on the law of privilege in any particular case.

Treat it as outsourcing, not as something new

Your firm almost certainly already has a process for letting a third party hold client material: a due diligence step, a written contract, a data processing agreement, a note on the file. An AI supplier sits in the same category. If the arrangement would satisfy you for a transcription service or a document hosting platform, the fact that the processing involves AI does not change the nature of the question.

What does change is the volume and sensitivity of what gets handed over. People give an AI tool whole files at once, including counsel's advice, internal risk assessments and candid client correspondence. That makes it worth doing the diligence properly rather than by analogy.

The questions to put to the supplier

Ask these before any live matter goes near the tool, and keep the answers:

• Where is our material held, and in which countries is it processed? Get the countries named, not a vague reference to the cloud.
• Who at the supplier can read it, in what circumstances, and is that access logged?
• Is our data kept separate from other customers' data?
• Will our material ever be used to train or improve a model that anyone outside our firm benefits from? You want a clear no, in the contract.
• Which subprocessors are involved, and will we be told before that list changes?
• How long is material retained, and what happens on deletion or if we leave?
• What happens to our data if the supplier is acquired or stops trading?
• What are the security controls on our own accounts: email verification, two-factor authentication, passkeys?
• How and how quickly would we be told about a security incident?

If a supplier cannot answer in writing, that is your answer. A tool that will not commit on retention and training is not one to put privileged material into, however good the output looks.

Consumer tools and personal accounts are the real exposure

Most of the privilege risk in a firm does not come from the system the partners chose. It comes from a fee earner at home pasting an attendance note into a free chat tool on a personal login, or using a browser extension nobody has assessed, because it saved twenty minutes.

In that situation the firm has no contract, no record of what was disclosed, no retention commitment and no way to answer a client who asks where their material went. If you later had to describe the circumstances in which a document was shared, you would struggle.

So the practical controls are mundane: approve a small number of tools, say plainly that client material goes nowhere else, make the approved route easy enough that people use it, and ask about it in supervision and file reviews rather than assuming the policy is being followed.

Keep privilege tidy inside the firm as well

Confidentiality is not only about outsiders. Two points worth checking in any tool you adopt:

Access by matter, not by rank. People should see the matters they are working on. Seniority alone should not open a file, particularly where you have information barriers in place for conflict reasons. Ask how the tool enforces that, and test it.

One matter per conversation. If a tool lets someone pull documents from several unrelated matters into a single thread, you have created a confidentiality problem of your own making. Keep the work for a matter in one place and keep other clients' papers out of it.

Also be careful about what goes into the question as well as what goes into the file. A prompt that quotes another client's facts for comparison has disclosed that client's material just as surely as uploading the document.

If you are asked about it

Clients, panel managers and insurers increasingly ask how AI is used and what happens to the material. Being able to answer calmly is part of the point of doing the diligence.

Keep a short, current note covering: which tools are approved, what they are used for, where data is held, the contractual position on training and retention, and who approved the arrangement. Check your engagement terms and privacy notice say something accurate about processors. Where your data protection obligations under UK GDPR require an assessment, do it and keep it. None of this is elaborate, but assembling it from scratch under time pressure, in response to a client's question, is unpleasant.

Where Alesis fits

Alesis is an AI assistant for UK law firms, made by L25 Limited, used through the web browser, with one conversation for a matter. A firm's information is held in the UK and processed only in the UK and the EU, kept apart from every other firm, and never used to train anything for anyone else. Inside a firm, people see only the matters they are on, and seniority alone grants no view. Every account proves its email address, and can add two-factor authentication and passkeys.