A law firm using AI still needs to know what information went into the tool and whether the answer can be relied upon. The SRA’s warning notice is a reason to examine those parts of the process.
The SRA’s Misuse of AI warning notice, published on 17 August 2026, focuses on inaccurate outputs and client confidentiality. It applies to firms and individuals the SRA regulates. Their responsibilities continue when they use AI. The notice does not require them to buy a particular product or appoint a consultant.
Describe the job the tool will support
For a first-pass chronology, specify the documents it will receive, what the draft should contain and how a person will review it. This gives the firm a task it can test and a way to identify which decisions remain with the responsible professional.
The SRA’s Risk Outlook report on AI in the legal market discusses opportunities and risks. A test of one defined task helps the firm assess a product’s suitability for that work. A successful demonstration does not establish that it suits every other legal task.
Check the source behind the answer
The warning notice highlights convincing but false information, including invented authorities. Where an answer refers to an authority, the responsible professional needs to locate and assess the original source. A citation attached to a fluent answer is not evidence that the proposition is correct.
For document-based work, ask for references alongside important statements, then open them. A letter’s date may differ from the date of the event it describes. Our legal prompt examples show how to request source references and keep unresolved points visible.
Establish what happens to the information
Check the product’s terms and settings before using matter information. Ask about processing, retention, access and any use for model training. Include connected services in those questions.
The Risk Outlook report identifies confidentiality and privacy as concerns. An initial test can use fictional documents with a reference answer. That lets the team investigate the task without entering client information. Removing a name alone should not be assumed to make a real file suitable.
Give someone responsibility for the review
Agree who checks the output before rollout. Users need to know what they can do and when to ask for help. The person responsible for the finished legal work needs to understand how AI has contributed to it.
The SRA’s effective supervision guidance describes an approach suited to the work, its risks and the experience of the people involved. Allow time for the review in the firm’s routine; assigning it on paper is not enough if nobody can carry it out.
Practise checking, then follow up
Use a fictional exercise that includes a mistaken date or an omission in a prepared answer. Ask the team to find and correct it against the documents. This tests more than their ability to produce a prompt.
Keep a route open for questions once the tool is in use. Record recurring errors and review whether the instructions, setup or task need to change. Include preparation and checking time when assessing productivity.
This article provides general information about AI implementation and links to the regulator’s guidance. It is not legal advice or an assessment of your firm’s compliance. Sources were checked on 26 October 2026.
If you need help turning these questions into a project, discuss your firm’s workflow and training needs with me.