The SRA's warning notice on the misuse of AI was published on 17 August 2026. The press release explained why: 42 reports of potential misuse between July 2025 and July 2026, some from senior judges, some from solicitors reporting themselves. Aileen Armstrong, the SRA's executive director for strategy and policy, put the principle in one sentence: "the use of AI does not change the professional standards expected of solicitors and law firms".
What the notice actually says
Four failure types, in its own words or close to them:
- Hallucinated "fictitious cases, references or seemingly factual assertions".
- False or incorrect citations in court documents "not checked or verified before submission".
- Confidential client information entered into public AI tools.
- Inadequate supervision of junior or non-authorised staff using AI.
And four requirements:
- "They must verify all content in legal submissions."
- "Reliance on an output of AI would not be a suitable defence."
- Client information "should only be entered into AI systems where appropriate" safeguards exist.
- Firms must have "effective governance structures, systems and controls in place to manage any risks, including those arising from use of AI."
It cites SRA Principles 1, 2, 4, 5 and 7 and a long list of Code paragraphs on competence, supervision, confidentiality and the duty to the court. It does not use the phrase "AI policy" and it does not mandate training; the effective supervision guidance, updated in June 2026, is where the supervision expectations for AI-assisted work live: outputs "subject to appropriate human review, scrutiny and professional judgement", and "an authorised individual retains ultimate responsibility".
What the Law Society adds
Generative AI: the essentials, updated 30 June 2026, is the practical companion. Its checklist: follow your firm's IT, AI, confidentiality and data governance policies; communicate the use of generative AI tools with the client; review outputs for accuracy; document inputs, outputs and errors. It is explicit that free online tools should get no confidential data, that real client data should never be used for testing or templating ("Always create and use fictional data"), and that training in ethical use is as important as training in the tools. On client disclosure it is candid: the SRA has no specific guidance, so the firm and the client should decide together, and the firm should communicate "when and how AI tools are used in their matters, where appropriate to do so".
What that means a small firm can do now
Read together, the two documents describe a firm where AI is used and a person stays responsible. That rules out very little and shapes everything. In practice:
- Enquiry handling and intake can be automated end to end, because no advice is given and every step is either factual or approved by a fee earner. This is where most firms should start, and the intake guide sets it out step by step.
- Attendance notes, client updates and chasers can be drafted by AI from the call or the file and sent by the fee earner. The review is the control.
- Document drafting from the firm's own precedents is fine as a first draft; anything with a citation is verified against the source before it leaves the firm. The document automation guide draws the lines.
- Legal research with generative tools is a starting point, never a source. The 42 reports are mostly this failure.
- Confidential and privileged material goes only into tools on a register, on business terms that prohibit training, with the decision recorded.
The three things to do this month
- Write the policy down. The notice requires governance and controls; the Law Society assumes a policy exists. We published an AI policy template for law firms built around both documents.
- Put the verification step on the file. Every citation in a court document checked against the primary source, and the check recorded, before filing. Make it a file requirement, not a reminder.
- Add the AI paragraph to your supervision arrangements. Who reviews AI-assisted work by trainees, paralegals and support staff, and how.
Then build, in the order above. The AI and automation for law firms page describes what we build for firms of 5 to 25 fee earners, the systems we build inside, and what it costs.
Founder, Anco AI. Builds the systems this site talks about, and runs his own business on them.
