This is general information, not legal or regulatory advice, and Graft is not a law firm. For advice on your obligations under the SRA Standards and Regulations, speak to your COLP, COFA, or a qualified solicitor.
Shadow AI is the AI your firm does not know about. A fee-earner drafts on a personal ChatGPT account. A paralegal runs a document through a free summariser to save an hour. A trainee pastes a client letter into a chatbot to "tidy up the wording." None of it appears in any system you control, and most of it never gets mentioned. That is the problem. You cannot govern, cost, or defend a tool you cannot see.
The SRA's warning notice on the misuse of AI, published on 17 August 2026, put confidentiality at the centre of the risk. Shadow AI is where that risk actually lives, because the unsanctioned use is the use nobody reviews.
What shadow AI is, and why it is not going away
Shadow AI is the same pattern as shadow IT, moved to a new tool. People adopt what helps them do the job, whether or not it has been approved. The reason it spreads inside firms is simple: for a lot of routine drafting and summarising, a general-purpose model genuinely saves time, and the person using it feels the benefit immediately while the risk stays invisible.
The scale is not marginal. One 2024 survey of 6,000 knowledge workers across the UK, United States, and Germany, run by Software AG, found half were using AI tools their employer had not issued. A separate global study by Microsoft and LinkedIn put the share of AI users bringing their own tools to work at 78%, and found 52% were reluctant to admit they relied on AI for their most important tasks. Those figures span every sector, not just law, but there is no reason to think solicitors are the exception, and the legal numbers point the same way: LexisNexis found 61% of UK lawyers were using generative AI in their work in 2025, up from 46% at the start of the year.
If most people using AI at work would rather not say so, the honest assumption for any firm is that some of it is already happening on client matters, and that the official picture understates it.
Why it matters for a firm specifically
For most businesses shadow AI is a data-governance headache. For a law firm it reaches the core professional duty of confidentiality.
The SRA notice quotes the Upper Tribunal's plain description of the danger: "to put client letters and decision letters from the Home Office into an open source AI tool, such as ChatGPT, is to place this information on the internet in the public domain." The regulator's own conclusion follows: "Using AI tools in this way will likely breach client confidentiality and as a result, legal professional privilege may be permanently waived and unable to be recovered." Once a client's file has gone into an open model, you cannot pull it back out.
This is not a hypothetical the regulator is bracing for. The SRA has said it received 42 reports of potential AI misuse between July 2025 and July 2026. Shadow AI is precisely the category that produces those reports, because it is the use that reaches a client matter without a check in front of it.
Why banning it makes the problem worse
The instinct is to ban it. Block the domains, send the all-staff email, declare public AI off-limits. It feels like control. It removes the wrong thing.
In the same Software AG survey, 46% said they would carry on using their own AI tools even if their firm banned them outright. Set that next to the 52% who already prefer not to admit they use AI, and the effect of a ban is not hard to predict. The usage does not stop. It moves onto personal phones and home laptops, where you have no log, no policy, and no way of knowing a client matter went into a public model until it surfaces somewhere you cannot retract it. A ban does not remove the risk the SRA is worried about. It removes your sight of it, and with it any chance of evidencing that you had controls at all.
Get visibility first, then decide
You cannot make a defensible decision in the dark. A blanket rule written without knowing where AI already sits either strangles the work that was helping or waves through the work that was dangerous, because you had no way to tell them apart.
So the first move is not a policy. It is a picture: where AI is already used across the firm, in which matters, and against what data. With that in hand, the decisions get straightforward. Shut down what is unsafe. Put sanctioned, private tooling around what is valuable. Keep a record that the controls exist and are working, which is exactly the oversight the notice asks for.
Where Graft fits
Graft connects to the systems your firm already runs on, your practice and document management, email, and finance tools, and reads how work actually flows. It surfaces the AI use you know about and the shadow AI you do not, mapped to the matters and the data each touches, without asking fee-earners to self-report on a form they will rush.
From there the loop closes the way the rest of Graft works: see where AI is used, judge each use on whether it is safe and whether it pays off, and put a number in pounds on the parts worth keeping. "Here is every place AI touches client work, and here is the evidence it is controlled" is a far better position than a policy nobody can prove anyone follows.
Graft's free connected assessment finds where AI is already at work in your firm by reading your real systems rather than asking. For the fuller picture built for firms, read AI for law firms, or how to respond to the SRA's warning notice without banning AI. The two risks that make shadow AI dangerous each have their own piece: whether ChatGPT is confidential for lawyers, and what happens when AI invents case law.