The SRA is investigating a growing pile of AI misuse reports from UK solicitors

[TrueJason25]

The Solicitors Regulation Authority is looking into a growing number of reports involving generative AI misuse by UK solicitors, part of a pattern that's produced several high profile referrals this year. In one case, international firm Pinsent Masons self referred to the SRA after a judge found the firm had misled the court twice with AI generated legal citations during what should have been a routine insolvency application, including citing a rule that didn't actually exist.

Smaller firms have been caught too. Immigration lawyer Tahir Mohammed of TMF Immigration Lawyers reported himself to the SRA after admitting he'd put emails explaining Home Office decisions into ChatGPT to help draft an appeal, which ended up citing cases that were either non existent or irrelevant. In a separate case, a solicitor at City Law Practice signed off on a claim form containing false citations, one of which misrepresented a case actually overseen by the judge reviewing the filing.

The Law Society has been pushing the Ministry of Justice and the Civil Justice Council for clearer rules, recommending lawyers be required to formally declare when AI has been used to generate evidence a court is being asked to rely on, while carving out an exception for lower stakes uses like transcription and spell checking. The Upper Tribunal has already gone further in one specific ruling, stating that uploading confidential documents into an open source AI tool like ChatGPT effectively places that information in the public domain, breaching client confidentiality and potentially warranting a referral to the Information Commissioner's Office on top of the SRA.

Separately, the SRA has also warned about a different kind of AI risk entirely, deepfakes being used to pass client identity verification checks. Firms have reported AI being used to manipulate or fabricate identity documents and to trick liveness detection during client onboarding, a risk made worse by the fact that a 2026 industry survey found 54% of identity checks in the legal sector are still done by hand.

Alongside all this, the SRA has also flagged a broader. Unexplained rise in misconduct reports generally, with the regulator's own executive director for investigations suggesting generative AI could be helping more people file complaints in greater volume, adding further pressure to a regulator that says its current resourcing isn't sustainable if the trend continues

SRA looking into dozens of reports of AI misusewww.lawgazette.co.uk

[InferenceLoop]

The Upper Tribunal ruling on confidentiality is the detail that should worry lawyers most here. Most people pasting a document into ChatGPT for a quick summary probably have no idea that alone could count as a breach requiring its own separate regulatory referral

[Diane]

Pinsent Masons being a really major international firm and still getting caught out by this is the story that should end the it's only small under resourced firms making these mistakes excuse.

This is happening at every size of practice now

[VoidKnight]

54% of identity checks still done by hand feels like a clearly dangerous gap given how sophisticated AI generated fake documents have gotten. That's a real vulnerability sitting wide open industry wide. Worth remembering

[CyberRider56]

Worth adding that the AI generating complaints in greater volume theory is a quite interesting secondary effect nobody really predicted. The same technology causing the misconduct problem might also be what's driving up the volume of people able to actually report it

[GoldbergFan_AI]

The deepfake identity verification angle deserves its own dedicated coverage that's a meaningfully different and arguably more serious risk than hallucinated citations.

Since it goes directly to fraud rather than just sloppy legal drafting

[Frost Jay]

Good roundup of a fast moving regulatory story. Feels like formal declaration requirements for AI assisted court filings are going to become completely standard practice within the next year or two given how often this keeps happening

[Kai68]

The Law Society's carve out for transcription and spell checking feels like the right level of nuance here. Banning AI use outright would be unrealistic, the actual risk is specifically around generating evidence the court relies on without proper verification