In partnership with
I suspect this week's most important story is also the least surprising. A new survey confirms what many of us have suspected for some time: most UK lawyers are using AI tools their firms have not approved, and firm leadership has no idea it is happening. Elsewhere this week, OpenAI has signalled its own legal AI play, another firm has been referred to the SRA over AI-generated fake citations, and an ambitious open-source legal AI system has gone live on GitHub.
AI in Practice
59% of UK lawyers are using AI their firms haven't approved
A survey of 200 UK legal professionals by Censuswide for Access Legal, published on 19 May, found that 59% of fee earners admit to using unapproved AI applications on client work. The tools in question include the free version of ChatGPT and other public platforms. Use is highest among paralegals (71%) and solicitors (57%).
The more revealing finding is the perception gap at the top. 68% of firm leaders said they had "full visibility" of AI use at their firm and believed they faced "zero risk" of unapproved AI use. Those two numbers cannot both be right. If six in ten fee earners are using tools the firm has not sanctioned, a leadership team claiming zero risk is not in control of its own compliance position.
The survey also points to a supply-side failure. Half of fee earners said they want AI built into their case management system, but only 25% of firms currently offer that. When the firm does not provide adequate tools, people find their own. That is entirely predictable, and the responsibility sits with the firm, not the individual practitioner who is trying to do their job more efficiently.
The risks are not theoretical. Uploading client material to a public AI platform raises serious questions about confidentiality and privilege, and the SRA's existing guidance already makes clear that firms must have appropriate systems and controls in place for any technology they use. A firm whose fee earners are routinely using tools that the compliance team does not know about has a governance problem, not just a technology problem.
Takeaways
Act: if your firm has not audited what AI tools staff are actually using (as opposed to what is formally approved), do so now. Anonymous internal surveys are a reasonable starting point.
Watch: the SRA's response. This survey, combined with the recent wave of AI-related referrals, creates obvious pressure for the regulator to issue updated guidance or begin targeted compliance reviews.
Risk: the gap between leadership perception and actual practice is itself a regulatory exposure. A firm that claims full control while its fee earners are using unapproved public tools is in a worse position than one that acknowledges the problem and acts.
Read: ResultSense | Today's Family Lawyer
On your radar
Another SRA referral over AI-generated fake citations: a judge has referred solicitors at AML Legal to the SRA after three documents submitted in support of a civil appeal application contained incorrect citations of authorities, suspected to have been generated by AI. The judge described the conduct as "inexcusable on the part of a professionally qualified lawyer" and called for judges to take a "robust approach" to what is now "a serious threat to the integrity of the justice system." This follows last week's Gordon & Thompson ruling, where HHJ Charman attributed the failure to firm management rather than the individual solicitor. The two cases together suggest that the judiciary is losing patience, and that both individuals and firms are in the firing line. Why it matters for UK lawyers: the direction of travel is unmistakable. If your firm uses AI for research or drafting, mandatory verification of all citations before filing is no longer optional. The question is not whether the SRA will act on these referrals, but when. (Law Gazette)
OpenAI plans "Codex for Legal": sources told Artificial Lawyer that OpenAI is developing a legal-specific AI offering, likely branded "Codex for Legal," as part of a broader strategy of vertical-specific tools across major business sectors. The product could take the shape of plugins (similar to Anthropic's approach with Claude for Legal, which last week expanded to 12 plugins and 20+ connectors). OpenAI is also building out a "Deployment Company" with forward-deployed engineers to help enterprise customers turn its models into practical tools. Why it matters for UK lawyers: the legal AI market now has three major foundation model providers (Anthropic, Microsoft, and OpenAI) competing directly for law firm business, alongside the established legal tech vendors. For firms still evaluating their options, the competitive pressure should drive better products and pricing, but it also means the market is moving faster than many procurement cycles can keep up with. (Artificial Lawyer)
Clients are driving law firm AI adoption decisions: a report from Artificial Lawyer, published on 21 May, found that more than half of law firms choose their legal AI tools based on client requests. Clients are increasingly pressuring firms to adopt specific AI products to save time and increase output. However, the report also notes that many in-house teams use AI less deeply than they claim, creating a gap between the demand they place on their firms and their own actual usage. Why it matters for UK lawyers: if your AI strategy is being shaped by what clients ask for rather than what your workflows need, the tail may be wagging the dog. Client-driven adoption can work well when firms integrate the tools into real workflows. It becomes a problem when the tool exists to tick a box in a pitch document and nothing more. (Artificial Lawyer)
ICO automated decision-making consultation closes 29 May: the ICO's consultation on draft guidance for automated decision-making is open for one more week. The central finding from the ICO's preparatory work (a report on AI in recruitment, published 31 March) was that most employers did not recognise they were making solely automated decisions. The ICO's position is that human involvement must be meaningful: the reviewer must have the authority, discretion, and relevant information to change the outcome, not merely endorse it. This feeds into the broader statutory code of practice on AI which the ICO is now required to prepare under SI 2026/425 (in force since 12 May). Why it matters for UK lawyers: if your firm or clients use AI in any decision-making process affecting individuals, this is worth responding to. The deadline is 29 May. (ICO)
Lawve AI partners with Anthropic to bring community-built legal skills into Claude: Lawve AI (previously Lawvable) has become an official connector in Claude's legal ecosystem, making its catalog of practitioner-authored Agent Skills searchable and usable from inside a Claude conversation. The skills are written by practising lawyers, in-house counsel, and legal technologists, each capturing a specific workflow for a particular area of practice, jurisdiction, or language. Users can search by topic or jurisdiction, inspect the full methodology before acting on it, and apply the skill directly within Claude. I have been testing Lawve's skills for some time, and the quality of the practice-specific workflows is a step above the generic legal prompts that circulate on LinkedIn. Why it matters for UK lawyers: this is what the Claude for Legal plugin architecture was designed for. Rather than relying on a single model to know everything about every practice area, the connector approach lets specialists publish their working methods and makes them available to the profession. For smaller firms without the resources to build custom AI workflows, browsing the Lawve catalog is a practical starting point. (Lawve AI)
Ad Break
In order to help cover the running costs of this newsletter, please check out the advert below. In line with my promises from the start adverts will always be declared.
Your prompts are leaving out 80% of what you're thinking.
When you type a prompt, you summarize. When you speak one, you explain. Wispr Flow captures your full reasoning — constraints, edge cases, examples, tone — and turns it into clean, structured text you paste into ChatGPT, Claude, or any AI tool. The difference shows up immediately. More context in, fewer follow-ups out.
89% of messages sent with zero edits. Used by teams at OpenAI, Vercel, and Clay. Try Wispr Flow free — works on Mac, Windows, and iPhone.
For Review
Lavern: a 67-agent open-source legal AI system (Artificial Lawyer / GitHub)
Antti Innanen, a well-known legal tech practitioner, has released Lavern as open source under an Apache 2.0 licence. The system contains 67 specialist agent prompts, eight workflows, and a debate protocol in which agents review documents, post findings with cited evidence, and run three layers of verification before pausing at a human gate for critical decisions. It took six months to build and runs to over 155,000 lines of code. It is not a law firm and does not provide legal advice, but it is worth examining if your firm has any technical capacity. The multi-agent verification approach, in particular, is a more thoughtful architecture than most of what passes for "legal AI" on GitHub.
Read or listen: Artificial Lawyer | GitHub repo
"Claude for Legal moves centre stage" (Artificial Lawyer)
A follow-up to last week's coverage of Anthropic's Claude for Legal launch, arguing that the product's deep integration with Microsoft 365 and its plugin architecture are positioning it as the default AI layer for law firms already in the Microsoft ecosystem. The piece suggests that Claude for Legal's approach of embedding directly into existing workflows (rather than requiring lawyers to switch to a separate tool) may prove more significant than the model's benchmark scores. Worth reading for the strategic framing, particularly if your firm is weighing up which provider to invest in.
Read or listen: Artificial Lawyer
Practice Prompt
Try the below prompt to audit your firm's AI tool usage and identify shadow AI risk. Ensure you fill in context and constraints and other aspects marked with {}. Remember to adhere to the Golden Rules and do not upload confidential or privileged information to public tools.
You are a legal compliance assistant. Your task is to help a UK law firm audit its exposure to shadow AI risk, following survey findings that 59% of UK fee earners use unapproved AI tools on client work while 68% of firm leaders believe they have full visibility and face zero risk.
Firm details:
- Firm size and type: {e.g., "6-partner high street practice" / "50-lawyer regional firm" / "200-lawyer City firm"}
- AI tools currently approved for use: {e.g., "Microsoft Copilot (enterprise licence)" / "CoCounsel for document review" / "none formally approved"}
- AI tools you suspect staff may be using without approval: {e.g., "ChatGPT free tier, Claude personal accounts, Google Gemini" / "unknown"}
- Current AI usage policy: {e.g., "informal guidance issued by email in 2024, no formal policy" / "formal policy prohibiting public tools but no enforcement mechanism" / "no policy at all"}
- Practice areas: {e.g., "commercial litigation, employment, corporate M&A"}
Using the framework below, produce a shadow AI audit for this firm:
1. **Discovery**
Design a short, anonymous internal survey (no more than 10 questions) to find out:
- Which AI tools staff are actually using (approved or otherwise)
- What types of work they are using them for (research, drafting, summarisation, client communications)
- Whether client material or privileged information has been uploaded to any public tool
- What is driving unapproved use (firm tools inadequate, not available, too slow to access, not aware of restrictions)
2. **Risk classification**
For each AI tool identified (or likely to be in use), classify it:
- Approved enterprise tool (data stays within the firm's infrastructure): note the contractual and technical basis for the confidentiality claim
- Public/open tool (data may leave the organisation): note whether use is formally prohibited, informally tolerated, or simply unknown
- Uncertain: list any tools where the data handling position is unclear
3. **Policy gap analysis**
Against the current AI usage policy:
- Does it explicitly distinguish between approved enterprise tools and public platforms?
- Does it specifically address confidential and privileged material?
- Is there an enforcement mechanism (e.g., blocking public AI URLs on the firm network, monitoring, training records)?
- Does it cover personal devices and personal AI subscriptions used for firm work?
4. **Remediation plan**
Produce a prioritised action list:
- Immediate (this week): steps to stop ongoing risk
- Short-term (this month): policy updates, staff communications, and training
- Medium-term (this quarter): technical controls, approved tool rollout, and audit mechanisms
- Ongoing: monitoring, periodic review, and how to handle future tool introductions
Constraints:
- {Add any firm-specific constraints, e.g., "We have no IT department" / "Several partners use personal ChatGPT accounts" / "We act for legally aided clients"}
- Apply the law of England and Wales throughout
- Do not invent regulatory guidance or case law
- This is a compliance planning exercise, not legal advice
How did we do?
Hit reply and tell me what you would like covered in future issues or any feedback. We read every email!
Thanks for reading,
Serhan, UK Legal AI Brief
Disclaimer
Guidance and news only. Not legal advice. Always use AI tools safely.
Recommended Newsletters
Below are a few newsletters that I recommend, for various reasons. Check them out!
