In partnership with
Two weeks after Microsoft put a legal plugin into Word, Anthropic has added a gallery of ready-made legal tools to Claude, including a cite checker and a contract review table,. We have also been reading what in-house counsel now expect law firms to tell them about AI, a reminder that the ICO's call for evidence on agentic AI closes in three weeks, and new research on how little people are told when a public body uses automation to decide their case.
AI in Practice
Anthropic's ready-made legal tools include a cite checker, but the duty to check stays with you
On 28 October Anthropic announced a gallery of what it calls legal "artifacts", described as "interactive tools, such as trackers and dashboards, that teams can use and share in Claude", as an extension of the Claude for Legal offering it launched in May.
Anthropic's own six are a cite check (which compares a document with its sources and flags mismatched quotations, wrong page numbers and unsupported claims), a jurisdiction survey matrix, a matter board built from engagement letters, emails and spreadsheets, a playbook deviation board, a regulatory horizon scan and a tabular review across a set of contracts.
Partner-built tools come from Goodwin (an AI governance benchmark), Harvey (an intake tool for business users), Foley & Lardner (an obligations checker) and PwC (a subpoena response tool), with more promised from Legora and Orrick. Anthropic says it has added more than 60 legal connectors since May, that "each answer or output has a link to its source", and that artifacts can be shared inside and outside an organisation, including with clients, using "sharing settings that ensure privacy and support ethical walls between teams" (Anthropic).
The announcement does not mention the UK or English law. The access-to-justice tools in the gallery are built on New York and District of Columbia law, the examples elsewhere refer to US states. The gallery sits under Claude Enterprise, and no pricing is given. For a UK firm the more useful way to read it is as a set of templates which someone at the firm would have to adapt, test and own before they are used on a client matter, in much the same way as a precedent bought in from elsewhere.
The cite checker deserves particular care, because it arrives in the same month that the SRA proposed writing the checking of AI output "by reference to primary resources" into the Statement of Solicitor Competence. A tool which compares a draft against the sources it is given will catch a misquoted passage or a wrong paragraph number, which is exactly the kind of error which sank the grounds of appeal in the Sarker case covered here on 16 October. It cannot, on the description given, tell you that a judgment it has not been given exists, that it remains good law, or that it says what the draft claims it says about English law, unless it is connected to a research source which can answer those questions. This author would treat it as a useful second pair of eyes within a supervised process, and would not record a document as checked on the strength of it alone.
The sharing features raise a separate point. A tool which can be shared with clients and across teams is convenient, but a matter board or contract table built from one client's documents is client confidential material, and the ethical wall is only as good as the settings someone at the firm chose. Artificial Lawyer's view is that the launch is "probably not so massive", since "Claude for Legal" is "really just Claude" plus legal skills and connectors, and this author would also pitch the excitement at about that level. The practical change is that firms already licensing Claude Enterprise now have more ready-made tools in front of their fee earners, and need to decide which of them are approved.
Takeaways
Act: If your firm licenses Claude Enterprise, decide which gallery tools fee earners may use, who adapts each one for English law work, and who may share artifacts outside the firm.
Watch: Whether UK-specific artifacts or English law connectors appear, and whether Legora's promised tools are aimed at UK practice.
Risk: A document recorded as "cite checked" when the tool only compared it with sources the fee earner supplied, or a client's material shared beyond the team entitled to see it.
Read: Anthropic
On your radar
In-house clients are asking which AI tools their law firms use: At a Gazette roundtable reported on 26 October (sponsored by Thomson Reuters), general counsel described asking external counsel which AI tools they use, wanting to know how a firm uses AI when issuing purchase orders, and expecting AI efficiencies to be "passed on in terms of cost, in a material way", while others raised privilege in AI tools and the effect of unchecked output on professional indemnity cover. One participant said they "always request zero data retention" from AI providers. Why it matters for UK lawyers: the question is moving from whether a firm uses AI to which tools it uses, on what terms and at what price, and it will come up at panel reviews and pitches. Prepare a short standard answer setting out the tools the firm has approved, where client data goes, how output is checked and how its use is reflected in fees, and keep it consistent with the AI wording in your client care letter. (Law Society Gazette)
The ICO's call for evidence on agentic AI closes on 20 November: Not this week's news, but not covered here before: the ICO opened a call for evidence on 8 October on how data protection law applies to agentic AI, with sections on data security, transparency, accountability, automated decision-making, fairness and purpose limitation, and lawfulness of processing. The responses will inform future guidance and the statutory code of practice on AI and automated decision-making which the ICO must produce under the Data (Use and Access) Act 2025. Why it matters for UK lawyers: agents which act across a firm's email, documents and case management system process client personal data in ways a chatbot does not, and the code will set the standard firms are measured against. Diarise the deadline (20 November), and if your firm is piloting agents, check that the DPIA covers what the agent can access and do on its own. (ICO)
In-house teams expect AI-generated correspondence to keep growing: Dentons' report From AI Adoption to AI Advantage, reported by Legal Futures on 28 October and based on interviews with senior in-house lawyers at Irish and international companies, found that almost every interviewee raised volume as a problem, because "polished, professional-looking text is now nearly cost-free", and that AI was mostly described as unlocking capacity rather than removing work. Partner David Kirton warned that genuine issues risk being "drowned out in the noise of AI-generated arguments". The findings are qualitative and the report gives no figures. Why it matters for UK lawyers: firms handling consumer complaints, employment grievances or debt disputes will recognise the pattern, and so may their own complaints teams. Check that your complaints procedure lets staff identify the actual complaint in a long AI-drafted letter and respond to that, within the time limits, without treating length as substance. (Legal Futures)
People are rarely told when automation helped decide their case: The Public Law Project's report Missing Pieces, funded by the Nuffield Foundation and reported by the Gazette on 26 October, found that only four of 15 automated systems believed to be in use by public bodies (in areas including social security, tax and adult social care) appear on the government's Algorithmic Transparency Recording Standard register, and that only one of 125 register entries was linked to an equality impact assessment. It concludes that the current "patchwork" of transparency rights does not give claimants enough to bring a judicial review, and calls for the register to be put on a statutory footing with enforceable rights to be notified. Why it matters for UK lawyers: for public law, welfare, immigration and community care practitioners, the existence of an automated tool may be the fact which grounds a challenge, and nobody is obliged to volunteer it. Add a request to your pre-action letters and subject access requests asking whether any automated or AI tool was used in reaching the decision, and check the ATRS register for the body concerned. (Law Society Gazette)
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.
You know the problem. Now build the fix.
You've spent years close to a broken industry or process. You know exactly what would fix it. But you’ve never had the time or resources to make it happen.
With Lovable, your expertise is all you need. Describe the solution to your problem — a website, a dashboard, an internal tool — and get professional-grade software back. You make every call on how it works, because you're the one who knows what it needs to do.
Start today and take $5 off your first month.
Discount of $5 applies to first month of any paid plan. You need to sign up via this webpage in order to claim the discount. The discount applies to new and returning users without an active paid plan. Terms apply.
For Review
SRA supervision and AI: what firms will be asked to prove (Eleonora Dimitrova, Legal Futures)
A guest post from 27 October arguing that a regulator looking at an AI failure will ask why the tool was chosen, what risks were assessed before it was adopted and who owned it, and that checking individual outputs will carry little weight as evidence of control on its own. Its list of records worth keeping (data-flow analysis, the contract review on confidentiality and jurisdiction, permitted and prohibited uses, a named owner, and a note of model updates) is a practical checklist to set against your own files, and it reads well alongside the SRA's competence consultation, which closes on 3 December.
Read or listen: Legal Futures
Anthropic increases legal vertical push with 'Artifacts' (Artificial Lawyer)
A short, sceptical read on this week's lead, which argues that a capable legal engineer could build similar workflows with existing tools and that the launch is mainly about keeping customers on Anthropic's models, in competition with Harvey, Legora, Thomson Reuters and LexisNexis. Worth reading before any conversation about whether the firm needs a specialist legal AI product or can work with what its general licences now include.
Read or listen: Artificial Lawyer
Practice Prompt
Try the below prompt to plan an acceptance test for a ready-made legal AI tool (such as a cite checker, contract review table or playbook deviation board) before your fee earners rely on it for English law work. 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 assisting a supervising partner, COLP or knowledge lawyer at a law firm in England and Wales. The firm has access to a ready-made legal AI tool, built by a vendor or a third party rather than by the firm, and wants to decide whether fee earners may rely on it for English law work and on what conditions. Your task is to produce an acceptance test plan and a short sign-off note. You are not assessing whether the product is good in general, and you are not concluding that its use is lawful, compliant or safe.
Context to apply:
- The tool and what the vendor says it does, from public material only: {e.g. a cite check tool which compares a document with its sources and flags mismatched quotations, wrong page numbers and unsupported claims; say "unknown" where not confirmed}
- Who built it and for which jurisdiction: {e.g. vendor-built, examples drawn from US law; say "unknown" if not stated}
- The sources or connectors it draws on, and whether it can check that an authority exists and remains good law, or only compare a document against material the user supplies: {describe, or say "not yet confirmed"}
- Sharing settings and who can share the tool or its outputs inside and outside the firm: {describe, or say "not yet checked"}
- What the vendor says about data retention, training on customer data and where data is processed: {summarise, or say "not yet checked"}
- The tasks the firm intends to use it for: {e.g. checking citations in skeleton arguments, checking quotations in witness statements, reviewing leases against the firm's playbook}
- Who will use it, by role, and who supervises them: {e.g. two associates, one trainee and a paralegal, supervised by the head of litigation}
- The firm's AI policy and approved tools list: {summarise, or say "none in writing"}
Produce the plan under these headings:
1. What the tool can and cannot establish
State in plain terms what the tool checks and what it does not, separating what the vendor has confirmed from what has been assumed. For a cite check tool, say expressly whether it can confirm that an authority exists, that it remains good law, and that it supports the proposition for which it is cited in English law, or whether it only compares the document against sources supplied by the user.
2. Test set
Design a test set of no more than ten documents or extracts from closed or non-confidential matters, with known errors deliberately included (for example a misquoted passage, a wrong paragraph number, a case cited for a proposition it does not support, an authority which has been overturned, and a citation to a case which does not exist). Describe each test item and the result a reliable tool should produce.
3. Pass criteria
Set out what the tool must catch, what level of missed errors is acceptable (if any), and which errors must always be caught by a human regardless of the tool's result.
4. Supervision and recording
For each intended task, state who checks the tool's output, against what primary source, and how the file records what the tool did and what the human checked. A document must not be recorded as checked on the strength of the tool alone.
5. Confidentiality and sharing
Set out the rules on who may share the tool or its outputs, with whom, and how ethical walls between teams are maintained. Identify any setting that should be switched off until it has been reviewed.
6. Sign-off note
Draft a short note for the supervising partner to approve, reject or limit use, with conditions, an owner and a review date, including a review whenever the vendor updates the tool.
7. What this plan cannot tell you
Flag every point resting on an assumption about the tool, the vendor, the firm's systems or the law, and say what the firm would need to establish to replace it with a fact.
Constraints:
- {Add firm-specific constraints, for example a client who has prohibited AI use, an insurer requirement, a court deadline, or a limited budget for testing time.}
- Apply the law of England and Wales, UK GDPR and the Data Protection Act 2018, and the SRA Standards and Regulations. Treat vendor descriptions as claims to be tested, not facts.
- Do not invent product features, vendor commitments, regulatory guidance, case law or statutory provisions. If you do not know what the tool does, write the question to put to the vendor rather than an assumption.
- Do not conclude that the tool is reliable, compliant or adequately supervised. The purpose is to test it and surface what is not yet known.
- Write in plain English and keep the whole plan short enough to read in five minutes.
- This is a planning aid, not legal, regulatory or compliance advice. The firm and its supervising solicitors remain responsible for any use of the tool and for the work done with it.How did we do?
Hit reply and tell me whether your clients have started asking which AI tools you use, and what you told them.
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!
