Microsoft Copilot for Legal: Who Controls the Future of Legal AI?

The central argument

The immediate disruption is not that Microsoft has made a machine capable of drafting legal text. It is that AI-assisted legal work is moving into the document, email and collaboration systems firms already purchase, govern and use every day.

That could lower friction and give smaller legal practices access to useful capabilities. It could also make it harder for independent specialists to compete on equal terms, deepen dependence on a single software ecosystem and tempt firms to confuse a familiar user interface with reliable legal analysis.

For family justice, the test must be whether the tools help people organise and evaluate information accurately, safeguard children, protect confidential data and improve the fairness of real proceedings — not merely produce legal-sounding documents faster.

1. What has Microsoft actually launched?

Microsoft’s 15 October announcement describes the general availability of more than 20 legal-specific skills. The capabilities build on an earlier Legal Agent preview and are now offered as legal plugins within Copilot in Word and Cowork. This is not merely the addition of a legal-themed prompt to a general chatbot.

Among the published capabilities are contract review against playbooks, drafting from approved precedents, clause analysis, comparisons between documents, tracked amendments, extraction of obligations and deadlines, structured memoranda and case chronologies. In Word, a user can review proposed edits as native tracked changes. In Cowork, some workflows span multiple documents, emails and other work material.

Microsoft also announced a growing network of external legal integrations. It identified providers across legal research, document and matter management, legal AI, e-discovery and agreement management. The partnerships are important, but the announcement makes clear that the partner-built plugins are beginning to roll out; it would be wrong to imply that every integration is already available in every organisation.

Availability is not the same as having Word

Microsoft’s official setup documentation specifies an active Microsoft 365 Copilot commercial licence for the legal plugin capabilities. It is therefore inaccurate to suggest that anybody who owns Word, or every Microsoft 365 customer, automatically has free access. Licensing, platform support, tenant settings and the availability of individual plugins still need checking.

Microsoft’s material is strongest when it explains concrete activities: reviewing a clause in context, proposing a tracked change, or connecting a relevant source. Those are testable functions. A general claim that a product can ‘do legal work’ is not.

2. Mike Murphy spotted the real commercial question

Mike Murphy’s LinkedIn commentary focused on a point that risks being lost beneath the launch publicity. Other major AI developers have already pursued the legal market, including through specialised products and integrations. Microsoft arrives with a different form of influence: its software is already embedded in daily professional work.

“The company that already owns the lawyer’s screen showed up last.”

— Mike Murphy, LinkedIn commentary on the Copilot for Legal launch, October 2026

It is a memorable way of describing the advantage of distribution. Legal technology providers traditionally had to persuade a practice to buy another system, integrate it with existing records, obtain internal security approval and convince lawyers to change their working habits. An incumbent productivity platform can offer new capabilities where many of those habits already exist.

That does not establish that Microsoft’s tool is better, or that every firm has already approved the legal plugins for confidential matters. Familiarity may reduce adoption friction; it is not evidence of legal competence or a complete compliance assessment.

Murphy’s question about whether specialist providers can still outperform the incumbent is the right next question. And I would add another: who benefits if the technology becomes the default way legal work is done?

3. Why Word matters: distribution can become a competitive advantage

Microsoft Word is not merely a blank sheet. In many organisations it is linked to Outlook, Teams, SharePoint, document retention practices, permission structures, templates and the systems used to move work between professionals. That familiarity matters because introducing technology is rarely just a procurement decision. It is a change in behaviour.

For a small firm, the opportunity is attractive. A document can be reviewed where it is drafted. A tracked change may be easier to inspect than a block of generated text copied out of a separate chat window. A properly configured search may locate material within approved organisational repositories. Potentially, that saves time.

But three different propositions must not be collapsed into one: the platform is easy to access; the platform is securely configured for this particular matter; the output is legally reliable. A provider can make the first proposition true without establishing the other two.

A test of substance, not just convenience

• Can the system identify the exact passage from which a factual proposition is drawn?
• Does it preserve dates, names, chronology and the status of contested information?
• Does it distinguish a clause, document summary or drafted argument from applicable law?
• Can a reviewer see exactly which words the AI changed and reject those changes?
• Does the organisation know which data sources and external plugins the system may access?
• Can the firm export its work and switch providers without losing essential records or workflow control?

The last question is commercially significant: a convenient system should not become a difficult system to leave.

4. Microsoft versus specialist legal AI is too simple a story

The launch announcement names specialist providers among Microsoft’s partners, including Harvey, Legora and DeepJudge, alongside document-management, research and e-discovery businesses. That creates a more complex competitive picture than a straightforward battle between Microsoft and independent legal AI companies.

An incumbent platform can both compete with specialist tools and distribute their capabilities. The value of a specialist may move towards proprietary knowledge, trusted content, domain-specific evaluation, superior workflow design, implementation expertise and the ability to prove quality.

The strategic question for those providers is whether access to customers remains genuinely open. If the default interface becomes the practical gateway to legal work, who controls which tools are discovered, what data is available to them, how users switch between providers and which products become economically attractive?

This is not evidence that Microsoft has restricted a particular legal AI partner or breached competition law. It is a framework for examining the incentives and potential market outcomes created by a powerful software ecosystem.

What should a buyer compare?

Neither column guarantees quality. The buyer needs evidence from their own legal tasks.

5. One day later: the CMA raises the wider competition question

On 16 October 2026, the UK Competition and Markets Authority announced a proposed strategic market status designation concerning Microsoft’s business software ecosystem. Its wider investigation began in May 2026 under the Digital Markets, Competition and Consumers Act 2024. The CMA is examining the consequences of Microsoft’s business software position for customer choice and competition.

The investigation raises issues including product bundling, interoperability, switching and the ability of rival AI providers to work effectively with Microsoft’s software. The October announcement is a proposed decision for consultation, not a final finding or designation. The CMA’s consultation is open until 13 November 2026, with the final decision expected in early 2027.

Microsoft has the opportunity to respond, and the CMA has not decided that any particular legal AI offering is unlawful. It would be both unfair and legally careless to turn a proposed regulatory designation into an allegation of established wrongdoing.

The relevant point for legal technology is broader. The CMA is looking at exactly the kind of ecosystem relationships that legal AI buyers now need to consider. Its official case page identifies a regulatory process focused on how businesses can exercise meaningful choice between products, including rival AI services.

6. The capabilities are real. So are the limitations.

Microsoft’s own Copilot for Legal transparency documentation does not claim that its outputs constitute legal advice. It warns that results may be incomplete or inaccurate, citations may be wrong, jurisdictional awareness is not automatic, and the legal skills were tested primarily in US English at release. It also says complex documents can contain nuance the system misses.

That disclosure should inform how firms evaluate the technology. The presence of a citation marker or a tracked edit is not proof that the legal proposition is correct. A source reference may point to a real clause while the inference made about that clause remains wrong.

Four kinds of error have to be separated

1. Source error: the wrong document, outdated precedent or irrelevant legal material is retrieved.
2. Extraction error: the system misreads a date, phrase, exception, page reference or person.
3. Reasoning error: it correctly describes a source but draws an unjustified legal or factual conclusion.
4. Workflow error: a correct suggestion is inserted, filed, circulated or acted upon without the required approval or confidential handling.

A vendor may report high retention of suggested redlines or favourable quality results from an internal evaluation. Such figures describe the vendor’s measurements and are not independent proof of performance in UK family proceedings or suitability for every type of legal instruction.

The question is not whether the product can write a persuasive paragraph. It is whether the organisation can identify the evidential and legal foundation for what that paragraph says.

7. What does any of this have to do with family justice?

Here the distinction between legal-sector marketing and the realities of the Family Court becomes particularly important. Microsoft’s examples are weighted towards documents, agreements, negotiation and commercial workflows. Those are important areas of legal work, but they do not represent the whole of justice.

A family case may concern alleged domestic abuse, coercive control, a child’s wishes and feelings, a disputed Cafcass account, social-work records, safeguarding information and several years of highly sensitive digital communication. A family practitioner has to determine what is alleged, what has been admitted, what has been found by a judge and what still requires investigation.

An AI-generated chronology may save hours of clerical work. But if it treats a police report as proof of an assault, an allegation as a judicial finding or a child’s reluctance as evidence of manipulation without adequate foundation, that convenience can do harm.

A hypothetical example: the apparently confident chronology

Imagine 500 pages of messages, school correspondence, court orders and safeguarding records. An AI system assembles a neat timeline. It accurately identifies several important dates but silently merges an allegation from an email, a second-hand summary in a professional report and a later court finding into one uninterrupted factual narrative.

The chronology looks coherent. It may even look impressive. Yet three different evidential statuses have been blurred. That is not a cosmetic defect: it risks changing how a reader understands the child’s experiences and the contested conduct of the adults.

In family proceedings, proper use demands traceable sources, precise labels, confidentiality safeguards and the willingness to say that a fact is disputed or not yet established.

The difference between a useful assistant and a substitute decision-maker

• Useful: identifying dated references within a lawfully accessible document set.
• Useful: extracting deadlines from a verified sealed order and flagging them for checking.
• Useful: comparing two versions of a draft statement without erasing important qualifications.
• Unsafe: treating the system’s summary as a verified account of abuse.
• Unsafe: automatically recommending contact arrangements or making a safeguarding conclusion from untested data.
• Unsafe: uploading confidential children’s records into an unsuitable service simply because the interface is available.

The core legal and welfare decisions remain matters for appropriately qualified professionals and the court, not a plugin.

8. Enterprise security is important — but it is not a permission slip

Microsoft says its legal capabilities operate within Microsoft 365 enterprise data-protection controls and honour existing document access permissions and sensitivity labels. Those are material safeguards, and organisations should compare them carefully with the protections offered by other vendors.

But the starting question for a UK family practice is not simply whether a platform is sold as ‘enterprise secure’. It is whether that particular organisation is legally entitled to process and disclose the particular information in that particular way.

A firm’s permissions may already be too broad. An AI feature able to draw on multiple files can surface information the user technically has access to but should not have been exposed to in practice. External plugins, retention periods, contractual terms, access logs and cross-border processing all require informed scrutiny.

The SRA’s August 2026 warning notice on misuse of AI specifically identifies inaccurate material in legal work and the risk to client confidentiality. The warning concerns regulated solicitors and firms, although the underlying confidentiality and data-protection questions are also important for others handling sensitive legal material.

Family Court information has its own restrictions

Proceedings concerning children may be subject to additional restrictions on communicating information. The relevant Family Procedure Rules and Practice Direction 12G need to be considered, along with the court’s orders and applicable statutory protections. A right to ask someone for support does not necessarily amount to permission to upload an entire case file to an external AI service.

The Information Commissioner’s Office guidance on AI and data protection also makes clear that personal-data processing by AI remains subject to the applicable UK data-protection framework. For a firm, that may require a documented assessment of lawful basis, contracts, security, risk, retention, access and the consequences for data subjects. A data protection impact assessment may be required where the proposed processing is likely to result in high risk.

9. Professional regulators and the judiciary are already setting boundaries

This launch does not take place in a regulatory vacuum. The SRA’s 17 August 2026 warning notice reinforces regulated practitioners’ responsibilities for accurate work and protection of confidential information. A tool’s convenience does not transfer professional responsibility to its manufacturer.

Equally significant is the Judiciary’s position statement on AI, published 9 October 2026. It describes a cautious but positive approach to AI in judicial work, accompanied by testing, governance and the retention of judicial responsibility for decisions. It says AI may assist, but must not replace judicial reasoning or the evaluation of evidence.

That statement is about the judicial use of approved systems. It does not mean Microsoft’s Copilot for Legal has been approved for judicial use or that professional users are subject to precisely the same technology approval process.

Nevertheless, the principles matter: verifiable outputs, appropriate oversight, confidentiality, controls and accountability. In any court-facing workflow, the human sign-off should be substantive, not a reflexive acceptance of something that sounds plausible.

It is not enough to say there is a human somewhere in the loop. The person must have the information, time, expertise and authority to challenge what the machine has produced.

10. Will this improve access to justice — or widen the gap?

For the large legal department that already licenses Microsoft 365 Copilot and has an information-governance team, the new capabilities may represent a relatively straightforward addition to familiar tools. But that is not the situation of many family court users.

A litigant in person may have a phone, a collection of screenshots, an unfamiliar court order and no budget for a specialist subscription. A small charity or affordable support service may lack the money or technical infrastructure to buy and safely configure enterprise tooling.

This creates a potential two-tier technology environment: increasingly capable, connected systems for institutions, and unverified consumer products for people already at a disadvantage. That outcome is not inevitable, but it should be part of the discussion whenever a provider claims to be transforming legal work.

Three questions about public benefit

• Access: Can a person without an enterprise licence obtain affordable, accurate assistance in understanding their procedural position?
• Accountability: Can the reader inspect the underlying source and challenge an AI-generated statement without specialist technical expertise?
• Value: Will saved time produce lower costs, clearer legal work or better outcomes for clients, or simply more volume and a different commercial margin?

If AI makes the preparation of a court document cheaper but does nothing to improve the accuracy of evidence, the fairness of participation or a child’s welfare, then the access-to-justice benefit remains unproved.

This theme connects with JSH Law’s earlier analysis, Legal AI Is Rebuilding the Legal Services Market. Today’s Microsoft launch, however, gives the question a concrete new dimension: distribution power inside the software people already use.

11. The JSH Law Legal AI Evaluation Test

Rather than asking whether Copilot, Claude, ChatGPT or a specialist provider ‘wins’, a legal practice should test each system against tasks that reflect the actual work it performs. The same approach can help organisations buying tools for family-justice support.

Test on authorised and appropriately anonymised datasets, with a known answer or source set where possible. Record errors as well as successes. Evaluate both the quality of the answer and the amount of skilled checking it requires.

The Six-Question Check for an AI-generated legal proposition

1. Source: What original document or authority supports it?
2. Status: Is it fact, allegation, professional opinion or legal interpretation?
3. Context: What surrounding detail could change the meaning?
4. Jurisdiction: Does it reflect current England and Wales law?
5. Risk: What harm follows if this proposition is wrong?
6. Responsibility: Which human is authorised and competent to approve it?

This is a JSH Law practical checking framework, not a statutory legal test or a substitute for a professional duty.

12. What should law firms and legal-support providers do now?

A launch announcement is a reason to evaluate, not a reason to rush confidential casework into a new workflow. Firms should begin with a defined use case and measurable evidence.

1. Inventory the tools already in use. Include approved Microsoft features, browser-based AI, specialist products, integrations and any informal ‘shadow AI’ practices.
2. Identify the work that merits automation. Consider draft comparisons, deadline extraction or administrative indexing before moving into complex legal judgment.
3. Confirm licensing and deployment. Check which legal skills and partner plugins are actually available and enabled in the organisation’s tenant.
4. Audit data rights and access. Review contracts, processor relationships, sensitivity labels, document permissions, confidentiality obligations and family-proceedings restrictions.
5. Run controlled evaluations. Use suitably authorised documents or synthetic cases to test sources, dates, jurisdiction, hallucinations and errors of legal inference.
6. Specify human review points. Identify who checks changes, signs off outputs, approves disclosure and acts on the result.
7. Monitor commercial dependence. Compare alternatives and confirm exportability, interoperability, cost and the ability to switch providers.

The same method can inform smaller legal-support operations, even if their tooling is modest. Strong evidence organisation, clear service boundaries and disciplined checking are more important than any particular software brand.

13. The question is not who owns the lawyer’s screen. It is what happens to the person outside it.

Mike Murphy is right to highlight the significance of Microsoft’s commercial position. Its new legal capabilities appear in a setting many professionals already know, and Microsoft’s distribution advantage is substantial. The arrival of legal-specific skills within that environment is a meaningful industry development.

But the legal profession should resist two easy narratives. The first is that a larger platform will inevitably provide the best legal analysis. The second is that specialist legal AI vendors are automatically displaced. Microsoft’s own partnership model suggests a more complicated market, while the CMA’s ongoing investigation shows that questions about choice and interoperability remain open.

For the people who come to JSH Law, the issue is more immediate. Can the technology help a parent understand what a court order actually requires? Can it keep disputed assertions separate from findings? Can it identify missing records without inventing evidence? Can it help someone facing domestic abuse prepare a clearer case without exposing them to additional risk?

These are not fringe questions. They test whether the promised gains in efficiency can be translated into justice.

The future of legal AI should not be measured by how quickly a document appears in Word. It should be measured by whether the resulting legal work is more accurate, more accountable and more accessible.

The major technology firms are entering legal work at scale. Now the legal profession — and the justice system — needs to decide what it expects from them.

14. Primary sources, credit and further reading

Credit and inspiration: Mike Murphy’s LinkedIn commentary on Microsoft’s Copilot for Legal announcement supplied the central market observation discussed in this article. His contribution is acknowledged with a short direct quotation; the wider legal and family-justice analysis is original JSH Law commentary.

• Microsoft — Introducing Microsoft Copilot for Legal (15 October 2026)
• Microsoft Support — Get started with Copilot for Legal
• Microsoft — Copilot for Legal transparency documentation
• Competition and Markets Authority — business-software proposal (16 October 2026)
• CMA — Microsoft’s business software ecosystem investigation and consultation
• Solicitors Regulation Authority — Misuse of AI warning notice (17 August 2026)
• Courts and Tribunals Judiciary — Position Statement on Judicial Use of AI (9 October 2026)
• Information Commissioner’s Office — Guidance on AI and data protection
• Family Procedure Practice Direction 12G — Communication of Information

Related JSH Law analysis

• Legal AI Is Rebuilding the Legal Services Market
• AI Liability in UK Legal Practice: Regulatory Duties and Risk
• AI Is Already in the Courts: Why Family Justice Needs Hard Lines
• JSH Law — Artificial Intelligence and Family Justice Hub

Need to make complex Family Court evidence more understandable?

JSH Law provides practical, evidence-led, non-reserved support to people preparing their own family cases in England and Wales. We help organise information, identify evidential gaps, structure statements and prepare for hearings — with careful attention to safeguarding, confidentiality and procedural fairness.

The work is agreed in advance and does not replace regulated legal advice or representation. We do not promise litigation outcomes or conduct reserved legal activities on behalf of clients.

Book a JSH Law Consultation →