AI compliance for law firms: client confidentiality first

For a UK law firm, AI compliance starts with confidentiality: before any client information goes into an AI tool, the firm must know where it is stored, who can access it, whether it trains the model, and what the contract says. The SRA's 2026 warning notice on misuse of AI and the ICO's AI guidance both put this first. Your COLP and DPO decide the rest.

Our pillar, AI client intake for law firms: a governance checklist, covers disclosure, refusal behaviour and the human gate for AI on the phone. And can a UK law firm let AI answer client calls? covers recording and consent. This post goes deeper on one question that runs across every AI tool a firm uses, not just intake: what happens to the client's confidential information. It describes published guidance; it is not legal advice.

What does the SRA say about AI and client confidentiality?

The starting point has not changed with the technology. Paragraph 6.3 of the SRA Code of Conduct for Solicitors says you keep the affairs of current and former clients confidential unless disclosure is required or permitted by law or the client consents. The SRA's compliance tips on AI also point firms to their obligation to keep client information securely.

What has changed is how directly the SRA now addresses AI. Its warning notice on misuse of AI, published on 17 August 2026, sets out several points that bear on confidentiality:

• It gives a pointed example: putting client letters and Home Office decision letters into an open source AI tool such as ChatGPT is, in the notice's words, to place that information on the internet in the public domain.
• It warns that both free and paid tools carry risk, because depending on the provider's terms, settings and technical architecture, information entered may be stored, retained or used to improve the tool.
• It says firms must ensure client data stays in a secure environment, is not accessed by unauthorised third parties, and is not used to train AI models except where explicitly authorised.
• It stresses that those who supervise remain accountable for work carried out through them, and that human oversight and professional judgement stay central.

The SRA's separate compliance tips on AI and technology add that firms should make clear when clients are interacting with an AI system, should give clients information they can understand about how technology is used, and should involve the COLP in overseeing compliance when new technology is introduced.

Read together, the message is not that AI is prohibited. It is that confidentiality obligations apply in full, and that the firm, not the vendor, is accountable for meeting them.

What does the ICO say about AI and personal data?

The ICO's guidance on AI and data protection is long and detailed. Three parts matter most for a firm choosing tools rather than building them.

• Transparency. The ICO says organisations need to be transparent about how they process personal data in an AI system, including purposes, retention periods and who data is shared with, and to give that information at the time it is collected.
• Security. The ICO's security chapter notes that AI systems can exacerbate known security risks, partly because data is copied, moved and stored in many places, and partly because AI systems depend on large amounts of third-party code.
• Impact assessment. The ICO's DPIA guidance says a data protection impact assessment is required where processing is likely to result in a high risk to individuals, lists AI among the innovative technologies that can contribute to that, and names large-scale processing of special category data as one of the cases that always requires one.

Law firms routinely handle special category data: health in personal injury and clinical negligence, criminal allegations, sexual life in family matters. Whether a DPIA is needed for a particular tool is for the firm's DPO or data protection lead to decide, but it is rarely a question that can be skipped.

If the AI vendor processes personal data on the firm's behalf, the ICO's guidance on controller and processor contracts applies too. Article 28(3) of the UK GDPR sets out terms that must be in the contract, including processing only on documented instructions, a duty of confidence, appropriate security, rules on sub-processors, assistance with data subjects' rights, end-of-contract provisions and audits.

Is confidentiality the same as data protection?

No, and treating them as the same is a common mistake. Data protection is about personal data: information about identifiable living people. Confidentiality is broader and older. It covers the client's affairs, whether or not they include personal data: a company's commercial strategy, the terms of a deal, the existence of a dispute.

An AI tool can be compliant with data protection law, with a signed processor contract and a completed DPIA, and still be a confidentiality risk if, for example, the vendor's staff can read inputs, or the vendor uses them to improve the service. The reverse is also possible. A firm needs to check both.

Legal professional privilege is a third, separate question, and whether a given use of a third-party tool affects privilege is a matter for legal advice, not for a checklist.

What should solicitors check before using an AI tool?

Ten checks, drawn from the SRA and ICO material above. They are questions to put to a vendor and to your own team, not a substitute for your COLP's and DPO's judgement.

• 1. Where is the data stored, and where is it processed? These are often different. A vendor may store records in the UK and send text to a model hosted elsewhere for processing.
• 2. Does anything the firm enters train or improve the model? The SRA's warning notice names this directly. Get the answer in the contract, not a sales email.
• 3. Who at the vendor can see inputs and outputs? Support staff, reviewers, sub-processors. Ask for the sub-processor list, dated.
• 4. How long is data retained, and can the firm change it? A retention period set for the vendor's convenience is not the firm's policy.
• 5. Is there a processor contract with the Article 28(3) terms? If the vendor processes personal data for the firm, the ICO's guidance lists what it must contain.
• 6. Is a DPIA needed, and has it been done? Decided by the DPO, before rollout, not after.
• 7. Which version are staff actually using? Many AI products have a consumer version and a business version with different terms. A firm licence does not stop staff using a personal account on the consumer version.
• 8. Is there a written policy on what may be entered? Which tools, for which data, by whom. The SRA's notice expects governance and training.
• 9. Are clients told? The SRA's compliance tips say firms should make clear when clients are interacting with AI. Check your privacy notice and, where relevant, your client care information.
• 10. Who supervises the output? The SRA is clear that accountability stays with the supervising solicitor. Name the person.

Where do law firms get AI confidentiality wrong?

The failures we hear about most are not dramatic breaches. They are small, everyday habits.

• Shadow use. A fee-earner uses a personal account on a free tool to tidy a letter, pasting in the client's name and circumstances. Nobody decided this; nobody knows it happened.
• Trusting the brand. A well-known vendor is assumed to be safe, without anyone reading the terms that apply to the specific plan the firm bought.
• Checking storage, not processing. The firm confirms records are held in the UK and does not ask where inference happens.
• No exit plan. Nobody has asked what happens to the firm's data when the contract ends.
• Policy without training. A policy exists, but staff have not been shown what it means in practice, so it does not change behaviour.

How should a firm start with AI safely?

Narrowly, and with the least sensitive data first. A sequence we see work:

• Inventory what is already in use. Ask staff, without blame, which AI tools they use and for what. The answer is usually more than the partners think.
• Pick one use with low-sensitivity data. Administrative tasks before client content; pre-client information before live matter files.
• Run the ten checks. Get written answers, file them, and involve the COLP and DPO before rollout.
• Write a one-page policy. Permitted tools, permitted data, who supervises, how to report a mistake.
• Train, then review. Show staff real examples. Review after a month and adjust.

For AI that picks up the phone, the first deployment described in our governance pillar applies: one office, one phone flow, pre-client data only, retention set by the firm, and a named person reviewing transcripts. Broader questions about AI in firms are covered in AI governance for law firms.

A worked example

This example is invented to illustrate the checks. It describes no real firm or vendor.

A twelve-fee-earner firm finds, from a staff survey, that five people use free AI tools for drafting. It adopts a business licence for one tool, and runs the ten checks. The vendor confirms that business inputs are not used for training, but the default retention period is longer than the firm wants and inference happens outside the UK. The DPO completes a DPIA, the firm sets a shorter retention period in the admin settings, and the policy restricts use to anonymised or non-client text until the DPO is satisfied with the transfer arrangements. Personal accounts are prohibited for firm work, with a clear route to request a new tool.

Nothing in that example required the firm to stop using AI. It required it to know what it was using.

How does WasilX approach confidentiality?

We would rather point you to the detail than summarise it in a sales paragraph. How we handle data is written up on our confidentiality page, and we expect firms to put the ten questions above to us as they would to anyone. What Cam, our AI Engagement Coordinator, does on the phone is limited: it takes the call at any hour, says it is AI, asks the firm's own qualifying questions, books the call back and sends the firm a summary. If you would like to go through the checks with your COLP and DPO in the room, that is the conversation we would suggest.

Frequently asked questions

Can solicitors use ChatGPT with client information? The SRA's August 2026 warning notice says putting client letters into an open source tool such as ChatGPT places that information in the public domain, and warns that free and paid tools may store, retain or use inputs. Whether any specific tool and plan is acceptable is for the firm's COLP and DPO to decide, on the terms that actually apply.

Does the SRA allow law firms to use AI? Yes. The SRA's guidance treats AI as something firms may use, provided they meet their existing obligations, including confidentiality, supervision and giving clients information they can understand.

Do we need a DPIA for an AI tool? Often. The ICO says a DPIA is required where processing is likely to be high risk, lists AI among relevant innovative technologies, and requires one for large-scale processing of special category data. Your DPO decides for each tool.

What should an AI vendor contract include? If the vendor processes personal data for the firm, the Article 28(3) terms listed by the ICO, plus clear answers on training, retention, sub-processors and where processing happens.

Is data protection compliance enough for confidentiality? No. Confidentiality covers the client's affairs, not only personal data. A tool can meet data protection requirements and still expose confidential information, so check both.