TL;DR: The SRA AI warning notice, published on 17 August 2026, tells solicitors and firms that using AI does not reduce their professional responsibility. It flags two problems: false information in court documents, and client data ending up in public AI tools without proper safeguards. If your firm has no policy on this yet, you must write one as soon as possible.
What Is The SRA AI Warning Notice?
The SRA published a warning notice on 17 August 2026 about the misuse of AI in legal services. It applies to every firm and solicitor the SRA regulates. The notice reminds you that your existing obligations under the SRA Principles and Codes of Conduct still apply, whatever tool produced the work.
Why Did The SRA Issue This Warning Now?
According to Legal IT Insider, the SRA received 42 reports of potential AI misuse between July 2025 and July 2026. That was enough to trigger a formal notice rather than a quiet word. If your firm treats AI use as a personal choice for each fee earner, that gap is what the SRA now expects you to close.
What Are The Two Main Risks Flagged?
The notice comes down to two core problems, AI hallucinations and confidentiality concerns.
What Counts As An AI Hallucination In Legal Work?
An AI hallucination is when a tool generates a case, a citation, or a fact that sounds real but does not exist. The notice points to four recent cases where this caused real problems in court.
Ayinde v Haringey LBC
Fabricated citations reached the court in this case, leading to a wasted costs application and a referral to the regulators.
UK v Secretary of State for the Home Department
This tribunal case involved inaccurate AI-generated authorities, and also raised the confidentiality point covered below.
BCP v A Mother
An unregistered barrister included AI hallucinations in her submissions. The court found no intent to mislead, but felt she underestimated how serious that was.
Cork and another v Smith
The High Court again set out the responsibilities on lawyers when errors trace back to unchecked AI output.
How Does AI Put Client Confidentiality At Risk?
In the UK v SSHD case, the tribunal compared putting client letters into an open tool such as ChatGPT to putting that information online. It becomes permanent and public. Privilege can be waived with no way back. This is not limited to free tools. Paid products can carry the same exposure, depending on their contractual and technical safeguards.
What Happens If Your Firm Ignores The SRA AI Warning?
The SRA has said plainly that firms failing to have proper regard to the notice risk disciplinary action. It ties this to Principles including honesty, integrity, and acting in clients’ best interests. It also points to supervision duties. If a junior colleague submits a hallucinated citation and nobody reviewed it, the supervisor can also be found in breach.
What Should Your Firm Do?
You do not need to stop using AI. You need a way of using it that you can defend if the SRA ever asks.
How Can You Reduce The Risk Of AI Hallucinations?
Every citation, case reference, and factual claim generated by AI needs a human check before it goes near a client or a court. Build a sign-off step into your process where someone other than the drafter checks each AI-assisted citation.
How Can You Protect Client Confidentiality When Using AI?
Client information should never go into a public AI tool your firm has not reviewed and approved. Know whether the tool stores data, trains its model, or lets others access it. Labyrinth Technology’s managed security services keep client data inside environments where those answers are known. Authentication solutions control who can reach that data in the first place.
How Should You Roll Out AI Tools Across Your Firm?
Ad hoc AI use, where everyone picks their own tool and works out the risks alone, is no longer defensible. Firms need vetted tools, trained staff, and someone checking the rules are followed.
Matt Dunn, our COO and CTO, puts it bluntly:
“Train your fee earners. Regularly. Don’t hide behind a policy. A one-off training session doesn’t cut it.”
They need to know which tools are approved, how they can be used, and what their responsibilities are to check the output.
Labyrinth Technology’s AI implementation and training service replaces scattered AI use with a governed setup you can point to if the SRA ever asks.
The SRA already holds 42 reports of AI misuse on file. One fabricated citation with your firm’s name on it would make that 43.
