Guide

Claude for small law firms: confidentiality, consent and the first skills

Before client matters go into Claude, ABA Formal Opinion 512 asks a lawyer to evaluate, for any AI tool, who inside and outside the firm could see the client's information, to read and understand the tool's terms and privacy policy, and to get the client's informed consent before putting their information into a tool that learns from it, where a boilerplate engagement-letter clause isn't enough. Anthropic says Claude Team isn't trained on chats by default, which is one fact for that risk check, not a reason to skip it; whether a given tool needs consent is the firm's call under its state's rules. The firm's methods, such as the NDA first pass and the contract summary, belong in skills with no client data in them, which AI Skill Pod shares to every lawyer's and paralegal's Claude, Cursor or ChatGPT (Pro, Business, Enterprise) from one invite link.

Short answer

  • Plan: Claude Team or Enterprise, no training by default.
  • ABA 512: a risk check for every tool; consent for self-learning ones.
  • Boilerplate consent isn't enough.
  • Skills: methods, never client files.

What ABA Formal Opinion 512 asks

  • Risk check for any tool: before client information goes into any AI tool, evaluate who outside and inside the firm could see it; the opinion says this depends on the client, the matter, the task and the tool.
  • Competence: have a reasonable understanding of the tool's capabilities and limitations, and keep it current as tools change.
  • Confidentiality: because many self-learning tools could reveal what goes in, get the client's informed consent before putting information about the representation into one.
  • Real consent: a general clause in an engagement letter isn't sufficient.
  • Terms: read and understand the tool's terms of use and privacy policy, or rely on someone who has.
  • Supervision and fees: managing lawyers set firm measures for lawyers and staff, and a lawyer billing by the hour bills only the time actually spent, including time reviewing the tool's draft.

From ABA Formal Opinion 512, 29 July 2024, read 5 November 2026, 23:32 JST, from the Wayback Machine's copy. State bars have issued their own opinions; follow yours. Not legal advice.

A worked example: a five-lawyer firm

Illustration: a made-up business. A five-lawyer business firm wants to use Claude for contract work. The managing partner:

  1. Reads Anthropic's commercial terms and privacy articles, records the risk check for Claude Team under ABA 512, moves the firm to Team, and switches off Rate chats.
  2. Decides with the partners which matters need a client's informed consent before any client document goes in, and writes a plain-language explanation for those clients.
  3. Writes the NDA first-pass and contract-summary skills from the firm's own playbook, with no client names, and puts them in the firm's pod so the two contract paralegals use the same method.

AI Skill Pod is this site's product. You can start on Free at $0, with no card, for up to 3 people and 10 skills; after the 14-day trial, your team stays on Free unless you buy Team ($5 per seat a month from 3 seats) or Business.

The first skills for a firm

  1. NDA first pass: each clause green, yellow or red, quoted, with the change to ask for.
  2. Contract summary: parties, term, renewal, termination, liability and the dates that matter.
  3. DPA checklist: roles, subprocessors, breach notice, transfers, deletion and audits.
  4. Meeting minutes: decisions, owners and dates, and what stays out of the notes.

Next step: write the first skill

The firm's own playbook is the best source for a skill. How to make your first Claude skill from a doc you have walks through one.

Skills for this

All skills

Questions

Can lawyers use Claude with client information?

That's for you to decide under your state's rules, not this page. ABA Formal Opinion 512 asks you first to evaluate, for any AI tool, who inside and outside the firm could see client information, and to get the client's informed consent before using a self-learning tool, where a boilerplate engagement-letter clause isn't enough. Anthropic says Claude Team and Enterprise aren't trained on your chats by default, one fact for that evaluation.

What does ABA Formal Opinion 512 require?

Issued on 29 July 2024, it says lawyers using generative AI must consider their duties of competence, confidentiality, communication, supervision and reasonable fees; read and understand the tool's terms of use and privacy policy; and get informed consent before putting client information into a self-learning tool.

What Claude skills should a small law firm set up first?

Methods that hold no client data: an NDA first pass, a contract summary, a DPA checklist and meeting minutes. Each marks what a lawyer must review, and none replaces that review.

AI skill

Have your AI review an NDA before you sign

Put this skill in AI Skill Pod and everyone on your team gets it in their Claude, Cursor or ChatGPT (Pro, Business, Enterprise): one invite link, free to start. NDA first pass has your AI give an NDA a first pass before a lawyer reviews it: each clause rated green, yellow or red with a quote from the text, the reason, and the change to ask for.

Use this skill in your AI

14 days free · No card needed

Sources

  1. ABA Formal Opinion 512: Generative Artificial Intelligence Tools (29 July 2024) — read at 23:32 JST from the Wayback Machine's copy (americanbar.org refuses automated reads): before client information goes into any GAI tool, lawyers must evaluate the risk that others outside or inside the firm could see it, a fact-driven analysis by client, matter, task and tool; lawyers using generative AI must consider competence, confidentiality, communication, supervision and reasonable fees (hourly billing covers only time actually spent); because many self-learning tools could disclose information relating to a representation, a client's informed consent is required before inputting it into such a tool, and boilerplate engagement-letter consent isn't enough; lawyers should read and understand the tool's terms of use and privacy policy; no consent is needed when no client information goes in (read 2026-11-05)
  2. Anthropic Privacy Center: Is my data used for model training? — read at 20:31 JST: commercial products such as Claude Team aren't used for training by default; thumbs up or down feedback keeps the conversation up to 5 years; Owners can switch off Rate chats (read 2026-11-05)
  3. Anthropic Privacy Center: How long do you store my organization's data? — read at 20:31 JST: deleted chats leave the history at once and the back end within 30 days (read 2026-11-05)