# AI in legal practice: the SRA's warning, citations and client confidentiality

*What the regulator now expects, why an unchecked citation ends careers, and what never goes into a public tool.*

## Production summary

- Modules to record: 2
- Total script: 897 words, about 6 minutes of finished audio
- Voices: Emma (host) and George (practice educator)
- Level: Solicitors, trainees, paralegals, legal executives, legal secretaries and practice managers

## Accreditation wording that must appear in the description

- **The CPD Certification Service** (planned): Application scheduled.
- **Solicitors Regulation Authority: warning notice on misuse of AI** (aligned): Written against the SRA's warning notice of 17 August 2026 and the SRA Principles and Codes of Conduct it cites. Our own mapping, with no endorsement from the SRA implied.

> Do not upgrade any of these words in a description or a thumbnail. Aligned is not accredited, and planned is not approved.


---

## The SRA's warning notice, and the citation that does not exist

**Runtime** about 3 minutes. **Words** 461. **Starts at** 00:00 in the full course recording.

### Learning outcomes to state on camera

- State what the SRA's warning notice says and its date
- Explain why reliance on AI is not a defence
- Describe what the court said in Ayinde
- Verify an authority before it goes into a document
- Explain the supervisor's exposure

### Script


`[CUE 1]` *A warning notice dated 17 August 2026 with its central sentence about accountability*

**EMMA**  [00:00]
George, the SRA's put something out on AI and everyone in my firm has a different version of what it says. What does it say?

**GEORGE**  [00:10]
It's a warning notice, called Misuse of AI, published on 17 August 2026. It isn't new law. It tells you how the regulator reads the existing Principles and Codes. And it says failing to have proper regard to it puts you at risk of disciplinary action.

**EMMA**  [00:28]
Has it banned AI?

**GEORGE**  [00:30]
No. It's about misuse. The core line is that you remain accountable for all work and advice delivered to clients, whether or not AI has been used in its preparation.


`[CUE 2]` *A court judgment with a citation stamped does not exist and an arrow to the regulator*

**EMMA**  [00:42]
And if the tool makes something up?

**GEORGE**  [00:44]
The notice says reliance on an output of AI would not be a suitable defence. That's the sentence to remember.

**EMMA**  [00:52]
What prompted it?

**GEORGE**  [00:54]
Cases. In Ayinde against Haringey, decided by the Divisional Court in June 2025, submissions contained case citations that were false. The court said lawyers remain personally responsible for the accuracy of what they put before a court, whatever tools were used.


`[CUE 3]` *A plausible looking case reference coming apart into name, court, year and citation, none of them real*

**EMMA**  [01:10]
And what did it do about it?

**GEORGE**  [01:13]
It said that where false citations are placed before the court, a reference to the regulator is likely to be appropriate. And the notice lists later decisions in the same vein during 2026, in the Upper Tribunal, the Family Court and the High Court.

**EMMA**  [01:30]
Is there a pattern?

**GEORGE**  [01:32]
A consistent one. The court doesn't accept that the tool was at fault. It treats minimising the problem as making it worse. And it refers.


`[CUE 4]` *Four checks on an authority: genuine, verifiable citation, relevant, advances the argument*

**EMMA**  [01:42]
Why do these tools invent cases at all?

**GEORGE**  [01:45]
They produce the most plausible text. Ask for an authority and you can get a believable name, court, year and citation, with a summary that supports you perfectly. None of it real.

**EMMA**  [01:58]
I could ask it to confirm.

**GEORGE**  [02:00]
Which proves nothing. And there's a subtler failure. It can cite a real case for a proposition the case doesn't support.


`[CUE 5]` *A supervisor reading a trainee's draft with the authorities open beside it*

**EMMA**  [02:09]
So what's the standard?

**GEORGE**  [02:10]
The notice sets it. All named authorities must be genuine, relevant, have a verifiable citation, and advance the argument. Four checks, and each needs the judgment itself.

**EMMA**  [02:21]
Give me the rule in one line.

**GEORGE**  [02:24]
Never cite a case you haven't opened. If you can't find the judgment on a recognised source, it doesn't go in.

**EMMA**  [02:32]
I supervise two trainees. Am I exposed?

**GEORGE**  [02:35]
Yes. The notice says those who supervise junior or non-authorised colleagues may also be found in breach if false citations reach the court without adequate review or supervision.

**EMMA**  [02:46]
What do I change?

**GEORGE**  [02:48]
Know whether AI was used in a draft. Check the authorities, not only the argument. And make it easy for a trainee to say I used a tool for this. One who's afraid to say so is your largest risk.

### Sources for the on screen credit

- Warning notice: Misuse of AI (17 August 2026), Solicitors Regulation Authority
- R (Ayinde) v London Borough of Haringey [2025] EWHC 1383 (Admin), High Court, Divisional Court
- SRA Code of Conduct for Solicitors, RELs, RFLs and RSLs, Solicitors Regulation Authority
- Generative AI: the essentials, The Law Society

---

## Confidentiality, privilege and what the firm must have in place

**Runtime** about 3 minutes. **Words** 436. **Starts at** 03:04 in the full course recording.

### Learning outcomes to state on camera

- Explain why client information in a public AI tool risks privilege
- State the safeguards required before client data enters an AI system
- Decide whether a given use of a tool is acceptable
- Describe what a compliance officer should have in place
- Record enough to justify a decision afterwards

### Script


`[CUE 1]` *A client file leaving a firm into a public tool, with a privilege seal breaking*

**EMMA**  [03:04]
George, the citations I understand. What's the other thing the SRA is worried about?

**GEORGE**  [03:10]
What you put in. Confidential client information entered into public or insufficiently secured AI systems. Paragraph 6.3 of the Code requires you to keep clients' affairs confidential.

**EMMA**  [03:20]
Surely pasting a paragraph into a tool isn't publishing it.

**GEORGE**  [03:24]
The notice says it is. Entering client information into a public AI tool is to place that information on the internet, in the public domain. The Upper Tribunal made the same observation in a 2026 decision.


`[CUE 2]` *Four safeguards around a client's data: secure, no third parties, no training, not kept*

**EMMA**  [03:39]
And the consequence?

**GEORGE**  [03:40]
A risk that legal professional privilege is waived. Privilege belongs to the client, and once it's lost you can't get it back. This isn't a data protection fine. It can decide the client's case.

**EMMA**  [03:54]
So when can client information go into an AI system?

**GEORGE**  [03:58]
Only where appropriate contractual, technical and organisational safeguards are in place. And the notice lists what firms must ensure.


`[CUE 3]` *Three questions on a card: approved tool, approved use, comfortable telling the client*

**EMMA**  [04:05]
Go on.

**GEORGE**  [04:06]
The data stays within a secure environment. It isn't accessed by unauthorised third parties. It isn't used to train AI models unless that's explicitly authorised and appropriate. And it isn't kept longer than necessary.

**EMMA**  [04:20]
How do I judge a tool on a Tuesday afternoon?

**GEORGE**  [04:24]
Three questions. Has the firm approved this tool, under a contract that gives those four protections? Is this use within what it was approved for? And would I be comfortable explaining to the client exactly what I entered and where it went?


`[CUE 4]` *A personal phone and a personal account with a client matter, marked as the worst combination*

**EMMA**  [04:40]
What if I take the client's name out?

**GEORGE**  [04:44]
That doesn't make a public tool safe. The facts of a matter usually identify the parties to anyone who knows of it. And the other side may well be one of them.

**EMMA**  [04:56]
Some people use their own account at home.

**GEORGE**  [05:00]
Which removes the firm's protections entirely. Personal account, personal device, client matter. That's the worst combination.


`[CUE 5]` *A firm's policy, training log and verification step, with a record showing what was checked and by whom*

**EMMA**  [05:06]
What should the firm itself have?

**GEORGE**  [05:08]
A written policy saying which tools may be used and for what. Training for everyone who drafts. A verification step for anything going to a court or a client. And a route to report a mistake without fear.

**EMMA**  [05:24]
Whose job is that?

**GEORGE**  [05:25]
The managers are responsible for compliance, and the compliance officer for legal practice must take all reasonable steps to ensure it.

**EMMA**  [05:34]
And records?

**GEORGE**  [05:34]
Paragraph 7.2. You must be able to justify your decisions and actions. If a document's challenged, showing what was checked, against what and by whom is the difference between a mistake and misconduct.

**EMMA**  [05:48]
And if I find a problem afterwards?

**GEORGE**  [05:50]
Tell your supervisor and your compliance officer at once. Courts have treated minimising the problem as worse than the error.

### Sources for the on screen credit

- Warning notice: Misuse of AI (17 August 2026), Solicitors Regulation Authority
- SRA Code of Conduct for Firms, Solicitors Regulation Authority
- Artificial Intelligence (AI): Guidance for Judicial Office Holders, Courts and Tribunals Judiciary
- Generative AI: the essentials, The Law Society

---

*Copyright WAJD Group. Built by WAJD AI.*