In partnership with
This week Anthropic launched Claude for Word in public beta, with legal contract review as its lead use case. Within days Microsoft announced its own legal-specific Copilot features for Word, including tracked changes and contextual comments designed for document-intensive professional work. Both companies are now competing directly to be the AI that sits beside lawyers in the application where most legal drafting actually happens. The practical implications for firms, and for the legal tech vendors that have built their businesses around document review, are worth thinking through.
The legal AI battleground moves to Microsoft Word
On 10 April 2026, Anthropic released Claude for Word in public beta, a native sidebar add-in for Microsoft Word on Mac and Windows, available through the Microsoft AppSource marketplace to customers on Claude Team and Enterprise plans. The launch was explicitly aimed at lawyers. Contract review is the first listed use case, with example prompts including summarising key commercial terms, flagging provisions that deviate from standard market position, making indemnification mutual, and working through reviewer comments as tracked changes.
The feature that matters most for day-to-day legal work is this: every change Claude proposes appears as a native Word tracked change, visible in Word's revision pane and reviewable exactly as a human collaborator's markup would be. Anthropic says the tool recognises multi-level legal numbering, defined terms, cross-references, and standard contract structures, and applies edits to individual clauses while leaving surrounding formatting intact. For any lawyer who has ever received a document back from the other side in a format that looks like it has been through a blender, that alone makes the tool worth testing!
Five days later, on 15 April, Microsoft announced new Copilot capabilities in Word that target the same audience. The update adds word-level tracked changes with a visible audit trail, contextual comments anchored to specific text, and automated table of contents management. Microsoft's demo centred on creating a due diligence report with risk assessment, flagging items for "legal sign-off," leaving little ambiguity about who the feature is designed for. Microsoft also confirmed that Anthropic's Claude models are now available inside Copilot itself as part of its multi-model approach, meaning lawyers using Copilot can access Claude's reasoning without leaving the Microsoft ecosystem.
The convergence here is worth a mention. For years, legal AI has largely been the province of specialist vendors building standalone platforms for document review, contract analysis, and legal research. What happened this week is that two of the largest technology companies in the world both launched legal-specific AI features inside the single application where most lawyers spend most of their working day. That changes the market. If a firm can get competent first-pass contract review from the AI already built into Word, the case for paying separately for a specialist tool becomes harder to make, at least for routine work.
There are, however, real limitations that firms should understand before getting carried away. Claude for Word has no access to a legal research database and cannot verify whether the cases or authorities it references are real, correctly cited, or still good law. It has no playbook enforcement (it cannot check a clause against your firm's standard positions across matters), no market benchmarking (it can flag that a clause looks unusual but cannot tell you whether it falls above or below market range), and no negotiation memory (each session starts fresh with no awareness of prior drafts or client preferences). Specialist legal AI tools have been building those features for years, and for complex transactional work they remain genuinely valuable.
Then there is the data retention question. Standard Claude Team plans do not include zero data retention, which means client documents processed through the add-in are handled by Anthropic's servers. For UK firms, the duty of confidentiality under SRA Principle 6 and the obligation to protect client information create a direct compliance question: has the client been told, and is the firm satisfied that the data processing arrangements are adequate? Purpose-built legal AI tools typically offer zero-retention agreements as standard. Firms trialling Claude for Word should check their plan terms carefully before uploading anything sensitive.
My thoughts
Whilst I have not yet tested either of these tools, I will set out my initial thoughts as this is an area I am specifically focussing my own firm's AI efforts. Last week this newsletter covered Harvey's autonomous Spectre agent and its vision of AI operating "like a team of associates" across entire client matters. That is impressive engineering, and for firms with the budget and the appetite it may well prove transformative. But for a firm like mine, and for the majority of UK practices that are not in the Magic Circle, the Harvey model is not an attractive path.
What matters is whether the AI you can actually get your hands on, inside the software you already pay for, does something useful on the work you are doing today. A Claude add-in that sits in Word and marks up a contract with tracked changes is closer to how most firms will actually adopt AI than a bespoke platform that requires a six-figure contract and a dedicated innovation team.
This writer has long argued that AI in legal practice will be won in the tools lawyers already use, not in new platforms they have to be persuaded to open. This week's announcements from Anthropic and Microsoft are the clearest evidence yet that the industry is moving in that direction, and that is good news for the profession as a whole.
Takeaways
Act: If your firm uses Word for contract work (which is to say, if your firm does contract work given Word's dominance), it is worth trialling both Claude for Word and the new Copilot features in a controlled setting with non-confidential documents. Understand what they can and cannot do before anyone uses them on live matters.
Watch: How the specialist legal AI vendors respond. If the basic contract review use case is now commoditised inside Word, the value proposition for standalone tools shifts toward playbook enforcement, market benchmarking, matter memory, and integration with practice management systems. Expect the pitch to change.
Risk: Data retention and confidentiality. Any firm deploying either tool on client work needs to be satisfied that the data processing terms meet the standard required by the SRA Principles and the firm's own information governance policies. Do not assume that because a tool is built into Word it has been vetted for legal use.
Read: Artificial Lawyer / Artificial Lawyer / Gavel / Claude Support
On your radar
Freshfields puts 5,000 staff on Google Gemini AI: Freshfields Bruckhaus Deringer has reported that more than 5,000 of its professionals now use AI tools built with Google's Gemini models, one year into the firm's strategic partnership with Google Cloud. Gemini has been integrated across multiple proprietary platforms, including the firm's Dynamic Due Diligence tool, its case management platform, and a multi-jurisdictional insights platform. More than 2,100 team members regularly use Google's NotebookLM Enterprise (with custom managed encryption keys) to synthesise complex materials and navigate large document sets. The firm also rolled out Google Workspace to approximately 2,800 users and has trained staff through an internal AI Academy delivered by 260 "AI Champions." Why it matters for UK lawyers: Freshfields are not simply saying "we use AI" but reporting specific adoption figures across named tools, which makes this one of the more certain accounts of Magic Circle AI deployment to date. The model of embedding AI across multiple internal platforms (rather than depending on a single vendor) and investing in structured training is worth studying, whatever your firm's size. (Legal IT Insider)
AI-generated client queries risk pushing up lawyers' fees: The Financial Times reported this week that UK law firms are seeing a surge in AI-generated letters, emails, and patent applications from clients, with the additional review time raising costs. Some firms are absorbing the burden on fixed-fee contracts; others are warning that hourly rates will need to reflect the extra work. One partner at a US-headquartered litigation firm described receiving so many AI-generated emails from a single client that the firm could not keep up with the volume. Why it matters for UK lawyers: this is the other side of the efficiency story that we reported on recently in a HR context. While firms invest in AI to reduce their own costs, clients are using the same tools to generate more correspondence, more queries, and more draft documents, each of which requires professional review. The irony is real: AI tools marketed as reducing legal costs are, in some cases, generating additional work for the lawyers meant to benefit. Firms on fixed-fee arrangements should consider whether their pricing models account for AI-inflated client output, and those on hourly rates should think carefully about how to communicate the additional cost to clients without appearing to penalise them for using technology. (Financial Times, 11 April 2026)
AI legislation may feature in King's Speech on 13 May: The government has confirmed it may introduce AI-specific legislation in the spring 2026 King's Speech, now scheduled for 13 May. If it appears, the bill is expected to be narrow, covering advanced AI models and AI-copyright provisions rather than broad AI regulation. This follows the government's decision to drop the proposed copyright opt-out (covered in this newsletter on 20 March) and the Lords committee's call for a more structured regulatory approach. Why it matters for UK lawyers: any AI bill, even a narrow one, will create new compliance obligations for firms advising clients that develop or deploy AI systems. It will also be the first piece of UK-specific AI legislation, which changes the regulatory landscape from "principles and guidance" to "statute." The fact that it remains uncertain whether the bill will appear at all is itself telling: the government is still weighing whether the political and practical conditions are right. Worth monitoring closely as May approaches. (Slaughter and May)
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 and actual products that I have tried, with some brief thoughts from me.
Want to get the most out of ChatGPT?
ChatGPT is a superpower if you know how to use it correctly.
Discover how HubSpot's guide to AI can elevate both your productivity and creativity to get more things done.
Learn to automate tasks, enhance decision-making, and foster innovation with the power of AI.
For Review
"Claude for Word for Lawyers: What It Does and What It Doesn't Do" (Gavel)
A clear-eyed assessment of Claude for Word's capabilities and limitations for legal work. The piece identifies three structural gaps that matter for contract practice: no market benchmarking (it can flag unusual clauses but cannot tell you whether terms fall within market range), no playbook enforcement (it cannot check against firm-specific standards across matters), and no negotiation memory (each session starts fresh). It also raises the data retention point, noting that standard Claude Team plans do not include zero data retention. Read it before deciding what role the tool plays in your workflows.
Read or listen: Gavel
"Understanding AI Hallucinations: Making Sure You Don't End Up at the Wrong Stop" (Above the Law)
Stephen Embry's piece examines why AI hallucinations are not random but follow a predictable pattern, tending to occur when answers involve novel or unclear legal issues where training data is sparse. The practical implication is straightforward: lawyers can anticipate where hallucinations are most likely and calibrate their review accordingly. If the question concerns a well-settled area of law, the output is more likely to be reliable; if it touches on emerging or ambiguous territory, the risk increases materially. A useful framework for anyone supervising AI-generated legal research or drafting.
Read or listen: Above the Law
Freshfields Google Gemini Integration: One Year On (Artificial Lawyer)
Artificial Lawyer's coverage of the Freshfields announcement provides additional context on how the firm structured its AI rollout, including the decision to deploy across multiple bespoke solutions rather than a single platform, the use of custom managed encryption keys for NotebookLM Enterprise, and the AI Champions training model. For firm leaders or innovation teams benchmarking their own programmes, this is one of the more detailed accounts of a large-scale legal AI deployment available.
Read or listen: Artificial Lawyer
Practice Prompt
Try the below prompt to draft a letter before action from a set of case notes or instructions. This is a first-pass drafting exercise, not a final product: the output will need a solicitor's review, particularly on limitation, the correct Pre-Action Protocol, and any without-prejudice points. 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 litigation solicitor at a UK law firm. Draft a letter before action based on the following instructions. The letter must comply with the relevant Pre-Action Protocol and the Practice Direction on Pre-Action Conduct and Protocols.
Client: {name, whether individual or corporate}
Proposed defendant: {name, address if known}
Area of claim: {e.g., breach of contract, professional negligence, debt, personal injury, property damage, boundary dispute}
Applicable Pre-Action Protocol: {e.g., Professional Negligence, Construction and Engineering, Debt Claims, Personal Injury, or "general PD on Pre-Action Conduct" if no specific protocol applies}
Summary of facts: {set out the key facts in chronological order, or paste in attendance note / case summary}
Losses claimed: {itemise the heads of loss and, where known, the amounts. If quantum is not yet fully particularised, say so and indicate that a schedule of loss will follow.}
Documents held: {list the key documents supporting the claim, e.g., contract, correspondence, invoices, expert report, medical records}
Any relevant limitation points: {e.g., date of accrual, applicable limitation period, whether the client is aware of any potential limitation defence}
The letter should:
1. **Identify the parties** and the capacity in which they are addressed.
2. **Set out the factual background** clearly and concisely, in chronological order, referencing the key documents where appropriate.
3. **State the legal basis of the claim**, identifying the cause(s) of action, the duty owed, the breach, and the causal link to the losses claimed.
4. **Itemise the losses** with as much detail as the instructions allow. Where quantum is not yet fully particularised, state what further information is being obtained (e.g., "a schedule of loss is being prepared and will be provided in due course").
5. **Comply with the applicable Pre-Action Protocol**, including:
- A clear summary of the facts on which the claim is based
- The legal basis for the claim
- The remedy sought (including the amount claimed, if quantified)
- The documents on which the claim is based (list or enclose copies as appropriate)
- The time limit for a response (typically {14 / 21 / 28 / 90} days depending on the Protocol)
- A reference to the relevant Protocol and a warning that failure to comply may be drawn to the court's attention
6. **Propose ADR** or at least indicate willingness to engage in alternative dispute resolution, as required by the Protocol and the Practice Direction.
7. **Address limitation** if there is any risk that the limitation period is approaching, without disclosing privileged advice on the point.
8. **Close with a clear deadline** for response and a statement of the client's intention to issue proceedings if the matter is not resolved.
Tone: firm, professional, measured. This is a letter from a solicitor, not a demand from a debt collector. Avoid aggressive language but be clear about the client's position and intentions.
Format:
- Headed "{Firm name}" or "Our client: {client name} / Your client: {defendant name}" as appropriate
- Paragraphs numbered for ease of reference
- "Yours faithfully" if addressed to the defendant directly, "Yours sincerely" if to their named solicitor
- Mark "Without Prejudice Save as to Costs" only if {the letter contains a settlement offer / the letter does not contain a settlement offer, in which case do not mark it}
Constraints:
- {Add any matter-specific constraints, e.g., "There is a confidentiality clause in the contract that may restrict what can be disclosed in open correspondence" or "The client wants to preserve the commercial relationship if possible" or "Limitation expires on [date] so the letter must go out this week."}
- Use UK legal terminology throughout (e.g., "claimant" not "plaintiff", "particulars of claim" not "complaint").
- Do not invent facts, authorities, or case law. Where the instructions are incomplete, flag what further information is needed rather than filling in the gaps.
- This is a first draft for solicitor review, not a final letter. Flag any points where the supervising solicitor should check the position (e.g., limitation, applicable Protocol, heads of loss, identity of the correct defendant).
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!
