The SRA’s warning notice and a recent Upper Tribunal decision make the same point from different directions: client confidentiality and legal professional privilege can be lost the moment client material is entered into the wrong AI tool.
Most firms did not decide to adopt AI. It arrived quietly, through the parts of practice that generate the most routine work. A fee earner pastes an enquiry into a chatbot to draft a holding reply. A trainee uploads a bundle to a summarising tool to get through it faster. Someone in the intake team uses a free assistant to tidy up a client’s account of events before it goes on file.
None of this feels like a regulatory event at the time. But the SRA and the Upper Tribunal have now both made clear that it can be, and that the consequences for confidentiality and privilege may be permanent.
What has the SRA said?
On 17 August 2026 the SRA published its warning notice on the misuse of AI. Among its concerns are instances of confidential client information being entered into public AI tools. The notice is clear that client information should only be entered into AI systems where appropriate contractual, technical and organisational safeguards are in place to protect confidentiality, and that “both free to use and paid for AI systems may pose risks to client confidentiality”.
It goes further on privilege, warning that legal professional privilege “may be permanently waived and unable to be recovered”. That is the point supervising partners should dwell on. The notice describes the consequences of sharing clients’ confidential information as irreparable, and a waiver of privilege, once it has happened, generally cannot be undone.
The underlying duty has not changed. Paragraph 6.3 of the Code of Conduct for Solicitors, RELs, RFLs and RSLs requires solicitors to keep the affairs of current and former clients confidential unless disclosure is required or permitted by law or the client consents, and paragraph 6.3 of the Code of Conduct for Firms places the same duty on the firm. What has changed is how easily that duty can now be breached in the course of ordinary work.
What did the Upper Tribunal say?
The judicial statement the SRA’s notice relies on comes from UK v Secretary of State for the Home Department [2026] UKUT 81 (IAC). The decision is best known for its treatment of false citations, but the Tribunal also addressed the conduct of an immigration adviser, who was also a solicitor, who had put client emails he had drafted into ChatGPT to improve them and had uploaded Home Office decision letters to summarise them for clients.
At paragraph 21, the Tribunal observed that putting such documents into an open source AI tool “is to place this information on the internet in the public domain, and thus to breach client confidentiality and waive legal privilege”. It contrasted closed source tools which do not place information in the public domain, which it said are available for tasks such as summarising without these risks. The decision’s headnote adds that such conduct might itself warrant referral to the SRA and should, in any event, be referred to the Information Commissioner’s Office. The Tribunal’s summary of conclusions at paragraph 60 puts the first point more generally, as a possible referral to the regulatory body.
Two points follow for firms. First, on the Tribunal’s reasoning, the damage is done when the material is uploaded, whatever the user intended. Second, the distinction the Tribunal drew is between tools that keep information within a controlled environment and those that do not. The SRA’s reminder that paid tools may also pose risks means firms cannot simply assume that a subscription equals safety. The question is what the terms and settings of the particular tool actually provide.
Which workflows need attention first?
Client intake
Enquiries often contain the most sensitive material a firm will ever hold, and they arrive before any conflict check or engagement. Staff using AI to draft replies or summarise enquiries may be processing a prospective client’s confidential information in a tool nobody has assessed. Intake may also engage paragraph 6.5 of the Code for Solicitors. That paragraph refers to current or former clients, but the SRA Glossary extends “client” to prospective clients where the context permits, so information received at this stage may affect whether the firm can act.
AI drafting and summarising features are increasingly built into email clients and browser extensions. The risk is that whole threads, including privileged advice, are passed to a third party service without anyone making a conscious decision. Firms should know which features are switched on across their systems, not just which tools staff have chosen to install.
Document review
Summarising bundles, disclosure and correspondence is where AI saves the most time, and where the volume of privileged material is greatest. This is exactly the task the Tribunal considered in UK. It is also where the case for a properly assessed, controlled tool is strongest, because the alternative is staff finding their own.
What should COLPs and supervising partners do?
The warning notice reminds firms that they must have effective governance structures, systems and controls to manage the risks of AI, and that COLPs must take all reasonable steps to ensure compliance under paragraph 9.1 of the Code for Firms. In practice, that points to a short list of actions.
- Find out what is actually in use. Ask fee earners, trainees and support staff which AI tools and features they use for intake, email and review. The answers are usually broader than expected.
- Approve tools for specific purposes. Keep a short list of approved tools and state, for each one, which categories of information may go into it. A tool approved for drafting generic correspondence is not thereby approved for privileged documents.
- Check the terms, not the marketing. For each approved tool, record whether inputs are used for training, how long data is retained, where it is processed and who at the provider can access it. If those answers are unclear, the tool is not suitable for client information.
- Supervise the use, not just the output. Paragraph 3.5 of the Code for Solicitors keeps supervisors accountable for work carried out through others. That includes how a trainee produced a summary, not only whether the summary is accurate.
- Have an incident route. Staff need to know that uploading client material to an unapproved tool must be reported to the COLP straight away, so the firm can assess its obligations, including any report to the Information Commissioner’s Office.
Firms should also keep a record of these decisions. If the SRA asks how the firm manages AI risk, a dated list of approved tools, the assessments behind them and the instructions given to staff is a far better answer than a general assurance.
What is the bottom line?
The convenience of AI is real, particularly in intake, email and document review. But both the SRA and the Upper Tribunal have now said plainly that confidentiality and privilege can be lost at the moment client material enters the wrong tool. For supervising partners and COLPs, the task is not to ban AI. It is to decide, in advance and in writing, which tools may handle which information, and to make sure everyone in the firm knows the difference.