AI ethics for lawyers: a short working guide

Five questions before each new AI use

1. Competence: do I understand what this tool does, what it gets wrong and how to check it?
2. Confidentiality: what goes in, where is it processed, and may the provider keep or learn from it?
3. Verification: who will check every fact and citation, and against what source?
4. Communication: would the client expect to be told, or is consent needed?
5. Fees: does my billing reflect the work actually done?

Where the rules live

Guidance is changing fast. Use the regulator's own site and confirm the date.

Worked example: approving a tool for first-draft summaries

A three-lawyer firm wants an AI tool to summarise long email chains. Working through the five questions: the lawyers test it on non-client material and learn it sometimes drops a date (competence). They read the provider's terms and find out whether inputs are retained or used for training, and decide to use only a business plan that says they are not, if that is what the terms state (confidentiality). They agree every summary is checked against the emails before it is relied on (verification). They check their engagement letter wording and regulator guidance on telling clients (communication). They agree time spent is billed as the time actually taken (fees).

Then they record the decision. A simple register row does the job:

Common mistakes

• Approving a tool once and never revisiting it after its terms change.
• Letting staff use personal accounts on consumer tools for client work.
• Checking that a citation looks right instead of opening the source.
• Billing AI-assisted work as if it took the time it would have taken by hand.
• Assuming a rule you read for one jurisdiction applies in yours.

Make it routine

Put the five questions in your AI policy and the approved-tools register. If you work in Notion, add a short checklist to the template for new AI uses so approval is recorded, not remembered. LawyerOS does not include compliance tooling; this is a page you can add.

• The ethics of AI tools in legal practice
• Firm AI policy template
• ABA Model Rules for cloud and AI

General information only, not legal advice. Check the rules of your own bar or regulator and your clients' terms.

Frequently asked questions

Do I have to disclose AI use?

It depends on your regulator, the court and the client. Check each.

Can AI replace legal judgment?

No. Duties of competence and supervision stay with the lawyer.

Is a free AI chatbot acceptable?

Consumer tools often have terms that are unsuitable for confidential information. Review before use.

Where should I record approval?

In a register kept with your policy; a Notion database works.

How often should I revisit an approval?

Set a review date when you approve it, and review sooner if the provider changes its terms or your use changes.