The Redline.Subscribe

The Resource · Updated June 2026

The Best AI Tools for Lawyers (2026): Tested for Safety & Confidentiality.

Most "best AI tools for lawyers" lists answer the wrong question. They rank tools by feature count and pricing as if you were buying a CRM. The real problem in 2026 is that most lawyers don't actually know which AI tools they can use with client information without creating a confidentiality issue — and the vendors making the loudest claims are often the worst offenders.

This page filters the market. Every tool below has been reviewed against three questions before we look at a single feature: does it train on your inputs, is a business agreement actually available, and would a competent supervising partner be comfortable signing it off for use on a live matter. We then cover what each tool is genuinely useful for, where it fits in a solo or small-firm workflow, and the conditions under which it should be kept away from client data entirely.

Editorial standard: no vendor pays for placement or for a rating change. Some "Visit" links may, in future, be affiliate links — they're marked accordingly and they do not influence which tools appear on this page or how they're scored.

How we evaluate

Three questions before any feature

  1. 01

    Does it train on your data?

    If a tool's defaults retain inputs to improve its models, it's not a candidate for client work — unless a business plan changes that contractually.

  2. 02

    Is a real business agreement available?

    A DPA (and, where relevant, a BAA) needs to be available on the SKU you actually intend to buy, not just on the enterprise tier you'll never reach.

  3. 03

    Would a partner sign it off?

    We test the tool against the standard a supervising partner would apply on a matter file review — not the standard a tech blog applies.

The shortlist

Tools, reviewed

Each tool is reviewed below with a confidentiality note, what it's best at, the pros that actually matter for a small-firm practice, and the conditions under which we'd advise caution.

Transactional drafting

Spellbook

Best for · In-Word contract drafting and redlining for transactional lawyers.

Trains on data: No

Confidentiality note

Spellbook states it does not use customer document content to train its models, and offers a signed DPA on paid plans. Document text is still transmitted to the vendor and (per their docs) to underlying model providers — review their current sub-processor list and data residency before sending privileged material.

Business agreement
Yes — DPA available
Trains on inputs
No

Pros

  • Lives inside Microsoft Word — no separate workflow to learn
  • Built specifically for contract redlining, not general chat
  • Vendor publishes a DPA and security overview suitable for firm review

Watch out · Verify the current sub-processor list and confirm whether your jurisdiction's data residency requirements are met.

Independently reviewed. We don't take fees in exchange for placement.

Visit Spellbook →

Drafting & analysis

Claude (Team / Enterprise)

Best for · Long-document review, contract analysis, and structured legal reasoning where you need careful, well-cited output.

Trains on data: No (on business plan)

Confidentiality note

Anthropic's Team and Enterprise plans contractually exclude your inputs and outputs from model training by default and offer a DPA. The consumer Claude.ai (Free and Pro) tiers are a different matter — avoid them for client content. Claude is particularly strong on long documents and tends to hedge rather than invent, but every output still needs your review.

Business agreement
Yes — DPA available
Trains on inputs
No (on business plan)

Pros

  • Excellent at long-context work
  • Tends toward caution and citation rather than fabrication
  • Team/Enterprise give a DPA and no-training-by-default

Watch out · The consumer Pro tier lacks the same protection — confirm you're on Team/Enterprise before any matter content goes in.

Independently reviewed. We don't take fees in exchange for placement.

Visit Claude (Team / Enterprise) →

Drafting & workflow

Microsoft 365 Copilot

Best for · Drafting correspondence, summarising matter folders, and working inside Word, Outlook and SharePoint.

Trains on data: No (on business plan)

Confidentiality note

Microsoft 365 Copilot operates inside your tenant's data boundary: prompts and responses aren't used to train the foundation models, and it inherits your existing Microsoft 365 compliance commitments. The caveat is configuration — web-grounded prompts and over-broad permissions widen what it can see. Confirm DPA terms and lock down permissions before enabling on privileged folders.

Business agreement
Yes — DPA available
Trains on inputs
No (on business plan)

Pros

  • Lives inside Word, Outlook and SharePoint
  • Enterprise data boundary keeps outputs in your tenant
  • Backed by an agreement many firms already hold

Watch out · Disable web-grounded prompts for privileged work, and review file-access permissions before enabling on matter folders.

Independently reviewed. We don't take fees in exchange for placement.

Visit Microsoft 365 Copilot →

General drafting

ChatGPT

Best for · Generic drafting, brainstorming, and learning — on the right tier, with no client identifiers.

Trains on data: Yes (on free / consumer plan)

Confidentiality note

ChatGPT Free and Plus retain inputs for model improvement unless you change a setting, and logs are kept for abuse review regardless — not for client content. Team and Enterprise contractually exclude inputs from training and provide a DPA, making them defensible for many drafting tasks once your firm reviews the terms. Same name, very different posture by plan.

Business agreement
Not for self-serve tier
Trains on inputs
Yes (on free / consumer plan)

Pros

  • Capable and familiar to most lawyers already
  • Team/Enterprise add a DPA and no-training by default
  • Fine on any tier for generic, no-client-data tasks

Watch out · Never put client-identifying content into Free or Plus. Treat consumer ChatGPT as a public tool.

Independently reviewed. We don't take fees in exchange for placement.

Visit ChatGPT →
Get all 52 prompts + confidentiality ratings in The Lawyer's AI Toolkit → £29Get the Toolkit →

Frequently asked

What lawyers actually ask us

Is it safe to use ChatGPT for legal work?
It depends entirely on which version of ChatGPT and what you put into it. The free and Plus tiers retain inputs for model improvement unless you change a setting, and OpenAI retains logs for abuse review even then — that is not an appropriate environment for content identifying a client or matter. ChatGPT Team and Enterprise contractually exclude customer inputs from training and offer a DPA, which makes them defensible for many drafting and summarisation tasks, provided your firm has reviewed the terms and decided which categories of matter content can be sent. Either way, ChatGPT is not a substitute for legal judgement and its outputs need to be verified before use.
What does the SRA say about AI?
The Solicitors Regulation Authority has not banned AI tools, but its published guidance and risk outlook make clear that solicitors remain responsible for the confidentiality of client information (Principle 7 and Outcome 6 of the SRA Standards and Regulations), the accuracy of any advice given, and competent supervision of the technology in use. In practice that means a firm should be able to evidence which AI tools it uses, what client data (if any) those tools see, what the vendor's data-handling terms are, and how outputs are checked. Guidance is updated periodically — always check the SRA's current AI / technology pages before relying on a specific position.
Does the ABA allow AI in legal practice?
Yes. The ABA's Formal Opinion 512 (July 2024) confirms that lawyers may use generative AI tools, provided they meet their existing duties — competence (Rule 1.1), confidentiality (Rule 1.6), supervision (Rules 5.1 and 5.3), candor to the tribunal, communication with the client, and reasonable fees. The opinion stresses that lawyers must understand a tool's data-handling practices before inputting client information and must independently verify any AI-generated work product. State bars have issued their own guidance on top of this; check yours for any additional obligations.