AI for law firms: privilege is not the first question
Lawyers tend to ask about privilege. The bar asks about confidentiality, and that is the one you will be held to.
Privilege can be waived by sharing a communication with a third party in a way inconsistent with keeping it secret. Whether typing a client's facts into an AI tool counts is still open in most courts. Nobody wants to be the test case.
Confidentiality under Rule 1.6 is broader. It covers all information relating to the representation, and Rule 1.6(c) requires reasonable efforts to prevent unauthorized disclosure of, or access to, it. That is the standard every state bar uses for AI, and the right lens for anyone looking at AI for small law firms.
This article is general information, not legal advice, and bar rules differ by state.
What happens to your words in a consumer tool
On the free or personal version of a chat tool, your conversation sits on the vendor's servers. Depending on the settings, staff may review it and it may train future models unless you turn that off. Defaults vary and have changed more than once.
Business versions usually differ. The contract typically bars training on your data, data is encrypted, and you control retention. Some vendors will sign a data processing agreement, and some a business associate agreement for health information.
The agreement matters more than the brand. A lot of AI for solo lawyers starts on a free account with a personal Gmail address. That is consumer terms, full stop.
Devices matter too. A chat history on a shared laptop or a phone with no passcode exposes a client file to anyone who picks it up.
The question is not whether you used AI. It is whether you read the terms before you pasted.
What the ABA and state bars say
ABA Formal Opinion 512, issued in July 2024, is the national reference point. It says to evaluate the risk that client information reaches the vendor or other users, to read the terms and privacy policy before putting client information in, and to get informed consent before using a tool that learns from what you type.
California, Florida, New York, Pennsylvania, New Jersey, the District of Columbia and others have their own guidance. The core is the same: know what the tool does with your data, keep confidential information out of tools that can expose it, supervise staff, and tell the client when it matters.
What to ask a vendor before you sign
Get written answers you can file. If a salesperson cannot answer these in writing, that is your answer.
- Do you train on anything I type or upload? Is that off by default?
- Who in your company, or among subcontractors, can see my data?
- Where is my data stored, for how long, and can I delete it permanently?
- Is it encrypted in transit and at rest?
- Will you sign a data processing agreement, and a business associate agreement if needed?
- Will you notify me if you receive a subpoena for my data?
- Do you have an independent security audit, such as a SOC 2 report?
- Can I turn off conversation history for the whole firm?
A one-page firm policy you can write today
You do not need a 20-page manual. A short policy everyone signs covers most of it.
- Approved tools. Name the one or two tools the firm pays for and has reviewed. Nothing else touches client work until a lawyer approves it.
- No consumer accounts for client work. Personal logins, free tiers and trials do not get client information.
- Strip what you can. Remove names, account numbers and dates of birth before pasting. A trust summary works just as well for 'Client A'. An immigration lawyer can draft a hardship letter without the A-number.
- Extra care for health, financial and children's information. Ask a lawyer first, and consider client consent.
- Tell the client. Put a plain sentence in your engagement letter that the firm uses AI tools under confidentiality safeguards.
- Devices. Turn off chat history where you can, lock every device, and skip shared computers.
If you already pasted something
Find out three things: which tool, under which terms, and what exactly went in.
On a business account with no training, exposure is low. Delete the conversation and note what you did. On a free account, delete it, turn off training and history, and ask whether the vendor will delete stored data. Then decide whether the client needs to know under Rule 1.4.
Privilege is probably not gone from a single paste, but do not guess. Most state bars run a confidential ethics hotline. Call yours.
How Shouldered handles this
HQ is an AI back office for law firms, and it works inside your firm's own tools rather than asking you to paste files into a new chat window. Treat it like any vendor: send us the eight questions above and get the answers in writing before anything client-related goes in.
Deborah at the Legal Desk keeps your approved tool list and tracks each vendor agreement's renewal and review dates. She prepares and tracks; she never gives legal advice, so whether a client needs to consent stays your call. Barnabas at the People Desk puts the policy on every new hire's onboarding checklist and logs who signed. See HQ for law firms.
Try this this week
- Open the settings of every AI tool you use, turn off training and history if you can, and screenshot the page.
- Read the data use and retention sections of your main tool's terms. Highlight anything about training, human review or sharing.
- If you use a free account for anything involving clients, price out the business version. For most small firms it costs less than one billable hour a month.
- Write the one-page policy above in your own words and have everyone reply that they read it.
- Add one sentence about AI use and confidentiality safeguards to your engagement letter template.
