By Mitch Edwards · · 6 min read

The SRA just put AI misuse in writing. Founders should read the warning too.

The SRA's new warning notice covers hallucinated court material and client confidentiality, after 42 reports of AI misuse in a year. The regulator's core line applies well beyond law firms: you remain responsible for your work, whether AI is used or not.

Yesterday the SRA published a formal warning notice on AI misuse. It received 42 reports of potential misuse in a year, and its message to solicitors is blunt: fail to have proper regard to this notice and you risk disciplinary action.

The notice is written for law firms. Founders should read it anyway, because both of the regulator’s concerns apply just as directly to anyone using AI on contracts and commercial documents. The consequences are simply commercial rather than disciplinary.

What the SRA actually said

The warning notice, published 17 August, names two concerns.

First, false and hallucinated material. Court and other documents containing invented citations, fictitious cases, and seemingly factual assertions with no basis in fact. The SRA has received reports from senior judges, and several solicitors have self-reported after relying on AI tools that generated inaccurate content.

Second, client confidentiality. Solicitors putting confidential client information into public AI tools without understanding what happens to it. The notice points to a 2026 Upper Tribunal decision that put it plainly: putting client letters into an open AI tool such as ChatGPT “is to place this information on the internet in the public domain.” Do that and legal professional privilege may be permanently waived. Not suspended. Waived, with no way back.

The line that matters most, from SRA executive director Aileen Armstrong: “Individuals remain responsible for the work they produce and the advice they provide, regardless of whether AI is used, or not.”

The notice itself goes further: AI has no separate legal personality. You cannot delegate accountability to a tool.

This sits on top of a growing case pile

The SRA notice didn’t come from nowhere. The Damien Charlotin database, which tracks AI hallucination rulings worldwide, has identified more than 1,900 cases. The sanctions have escalated with the volume: a roughly $110,000 fine in Oregon for a brief full of fabricated citations, the Nebraska Supreme Court indefinitely suspending a lawyer’s licence in April, and an Alabama family losing their trust appeal outright because their lawyer’s briefs relied on AI-fabricated citations.

Courts started with warnings, moved to fines, and are now ending careers and cases. The SRA notice is the UK regulator formalising the same direction of travel.

Why founders should care

You’re not filing court briefs. You’re doing two things that carry the same failure modes.

You’re signing contracts based on AI summaries. The hallucination problem the SRA describes is exactly what happens when an AI contract review invents a clause reference, misses a liability cap, or defaults to the wrong jurisdiction. Three patterns I’ve seen, anonymised and composited from a handful of similar situations:

  • The fabricated clause reference. A founder asked an AI tool to summarise an MSA. The summary referenced a “Section 12 (Limitation of Liability)” that put the cap at £100K. Section 12 in the actual contract was about IP assignment. There was no liability cap at all. The founder nearly signed.
  • The wrong jurisdiction. AI tools default to US framings unless told otherwise. A UK founder using an AI assistant to review a SaaS contract missed a restrictive covenant that would behave very differently under English law.
  • The confident wrong answer. AI doesn’t say “I’m not sure.” It says it with confidence. If you don’t know enough to challenge it, you don’t know enough to use it alone.

You’re pasting sensitive documents into public AI tools. This is the SRA’s second concern, and it applies to founders with almost no translation needed. If a solicitor putting client letters into a public AI tool is placing them in the public domain, a founder pasting a customer’s NDA-covered contract, a term sheet, or an employee dispute into a free AI tool is doing the same thing. Check what the tool does with your data: whether it trains on your inputs, where the data is stored, and whether your paid tier actually excludes training. Most free tiers don’t.

What founders should not use AI for without expert review

  • Contracts with uncapped or unusual liability provisions
  • Employment contracts or contractor agreements (especially restrictive covenants)
  • Investor documents (term sheets, SHAs, SPAs, EMI rules)
  • Customer MSAs and bespoke commercial deals over your standard size
  • Anything in a regulated sector (financial services, healthcare, legal, education)
  • Legal filings, formal legal correspondence, or anything going to a court or regulator

For everything else, AI is a fine first pass. The trick is knowing which bucket you’re in.

The trust-but-verify framework

I use AI on contracts every day. So do my clients. The trick is the workflow.

  1. Section by section. Don’t ask AI for an overall summary. Ask it to walk through each clause in order. Anything it doesn’t reference, you read yourself.
  2. Re-read every quote. If the AI quotes the contract, copy that quote and search the original document. If it’s not there word for word, the rest of the AI’s analysis of that section is suspect.
  3. Jurisdiction first, always. Before any review, tell the model which jurisdiction governs and which law applies. Make it state it back. If you skip this, defaults will bite you.

These three checks add 10 minutes. They’ve caught real errors in real contracts.

And following the SRA’s second concern, add a standing rule: sensitive documents only go into AI tools you’ve checked. Know where the data goes, whether it’s used for training, and who can see it.

The bigger point

The SRA’s core message travels well beyond law firms: you remain responsible for your work, whether AI produced it or not. Regulators are now saying out loud what courts have been enforcing for a year. As verification gets easier, not verifying gets harder to defend, for solicitors and for anyone running a business that depends on contracts.

The safest founders I work with treat AI as a fast first read. Never the final read.

Want help?

The Contract Confidence Kit bundles the playbook, the verification checklists, and the AI setup guide I use with founders. £97, instant download, 30-day money-back guarantee. It’s the cheapest insurance policy in your stack. Drop me a line at mitch@theedwardspractice.com or grab the kit from the Contract Confidence Kit page.

Sources

This post is general guidance for founders using AI to read contracts. It is not legal advice. The cases above are summarised from public reporting; the founder examples are anonymised composites. If you’re dealing with a specific contract or regulatory issue, get advice from a qualified lawyer.

Book a Discovery Call