AI Policies for Law Firms: A Framework PI Firms Can Use
AI Policies for Law Firms: A Framework PI Firms Can Use
AI policies for law firms, built for PI practice: the eight sections a policy needs, sample wording, and the bar duties each section answers.
A paralegal leans into your office. "Can I run this chart through the AI tool, or is that not allowed?"
You realize you don't know. Nobody does. Each person at the firm has made up their own answer.
That's what a missing policy looks like. AI policies for law firms exist to answer that question once, in writing, before it gets asked at 6 p.m. on a deadline.
I build AI software for PI firms, and I'd rather your team use it under clear rules than guess. Here's a framework built for PI work, with sample wording you can adapt.
What Should a Law Firm AI Policy Cover?
A law firm AI policy should name who owns it, which tools and accounts are approved, what client data may go into each, how output gets checked, and who signs off. It should also cover client communication, billing, training and a review date. For a PI firm, the data rules for medical records matter most.
That's eight sections. The rest of this post walks through each one, what the bar guidance says about it, and wording you can start from.
It's a starting draft, not legal advice. Adapt it with your own counsel and your state's rules.
The Myth: Pick a Safe Tool and You're Covered
Here's the comfortable version. "We bought the enterprise plan. It doesn't train on our data. We're fine."
The tool matters. It isn't the whole job.
Washington's bar tells a story that shows why.
In its 2025 advisory opinion, a senior partner named Anne rolls out a new AI research tool. The firm's IT team had picked a tool that protected client confidentiality and didn't train on outside data. So far, so good.
Then Anne sent "only a brief email introduction" and "left a more thorough training on the product up to the individual users." An associate used the tool on a motion without checking it.
Anne gave it a quick look and filed it. The judge pointed out "several key cases cited in the motion that are misquoted."
The opinion's diagnosis isn't about the tool. Anne "did not establish clear guidelines for the use of AI in legal work, including the need for human verification." Its advice: "It is best practice for the firm to have an AI use policy handbook as well as regular training for employees who utilize the product."
A safe tool used badly still produces a bad filing.
What Do the Bar Rules Ask For?
The bar guidance in our states points the same way. Supervising attorneys should set clear AI policies, train the people who use the tools, protect client confidences, and check output before it reaches a court. No rule we know of requires a specific template. What's required is that the firm's conduct meets the duties.
| Source | What it says about firm policy |
|---|---|
| California State Bar, Practical Guidance on Generative AI (May 14, 2026) | "Managerial and supervisory attorneys should establish clear policies regarding the uses of generative AI." Staff should "receive periodic training on the appropriate use of AI" |
| Same guidance | Supervisors "should review and update applicable firm policies, controls, and associated training" as AI changes |
| WSBA Advisory Opinion 2025-05 (Washington) | "Those using AI tools must receive adequate training and supervision." An AI use policy handbook and regular training are "best practice" |
| NYC Bar Formal Opinion 2024-5 (New York City Bar, Aug 7, 2024) | Lawyers "need guardrails and not hard-and-fast restrictions." Lists the duty to supervise lawyers and non-lawyers among the duties AI use touches |
If you practice in New Jersey, read the state's own AI guidance directly. We couldn't load the official text to quote it here. Bar guidance changes often, so check the current version wherever you practice.
A Policy Framework You Can Adapt
The most useful AI policies for law firms tie each rule to a duty. Each section below answers one.
The sample wording is deliberately plain. Change the names, the tools and the limits to fit your firm.
| Section | Sample wording | Duty it answers |
|---|---|---|
| 1. Owner | "[Name], managing attorney, owns this policy. Questions go to [Name] before the tool is used." | Supervision (CA Rules 5.1, 5.3; WA RPC 5.1, 5.3) |
| 2. Approved tools | "Only the tools and account types listed in Appendix A may be used for firm work. Personal accounts may not be used for client matters." | Confidentiality; vendor review |
| 3. Client data | "Client information goes only into tools marked 'approved for client data' in Appendix A. See the data table." | Confidentiality (CA Rule 1.6, B&P 6068(e); WA RPC 1.6) |
| 4. Verification | "Every citation is pulled and read before filing. Every figure in a demand is checked against the bill or record it came from." | Candor; competence |
| 5. Sign-off | "An attorney reviews and approves any AI-assisted work before it leaves the firm." | Supervision; professional judgment |
| 6. Clients | "The responsible attorney decides whether to disclose AI use to the client, and follows any client instructions limiting it." | Communication (CA Rule 1.4) |
| 7. Billing | "We bill actual time spent, including prompting and review. We don't bill for time the tool saved." | Fees (CA Rule 1.5) |
| 8. Training and review | "Everyone who uses an approved tool completes firm training first. [Owner] reviews this policy every [six] months." | Supervision; competence |
The verification line exists because of cases like Mata v. Avianca, where a federal court sanctioned lawyers for filing opinions ChatGPT invented. Our ChatGPT for lawyers guide has a five-step citation check you can paste into it.
The billing line follows California's guidance, which allows charging for "actual time spent," such as crafting prompts or "reviewing and editing generative AI outputs." Time the tool saved isn't time spent.
The Data Table Is the Heart of a PI Policy
A PI file is mostly sensitive data: medical records, bills, diagnoses, insurance details. So section 3 needs a table your staff can read in ten seconds. Here's a starting version.
| Data | Personal or free AI account | Firm-approved tool |
|---|---|---|
| Public law: statutes, published opinions | Allowed; verify against the official text | Allowed; verify |
| Firm templates and marketing copy | Allowed if no client facts are in it | Allowed |
| Client name, contact details, case facts | Not allowed | Allowed if approved for client data |
| Medical records, bills, diagnoses | Not allowed | Allowed if approved for client data |
| Settlement figures, liens, strategy notes | Not allowed | Allowed if approved for client data; attorney decides |
"Approved for client data" should mean someone read the vendor's terms against your confidentiality duty.
Vendors publish them; OpenAI's enterprise privacy page is one example of what to read. California's guidance says "reasonable efforts require more than reliance on generalized marketing assurances." Our HIPAA and AI demand letters guide covers the questions to ask a vendor.
Bad Policy Line, Better Policy Line
Most weak policies fail in the same way. They state a value instead of a rule. Nobody can check whether they followed a value.
Weak: "Staff must use AI responsibly and protect client confidentiality."
Better: "Client names, medical records and bills go only into the tools marked 'approved for client data' in Appendix A."
Weak: "Always double-check AI output."
Better: "Before filing, the drafter pulls every cited case from a research database, confirms the quote is on the cited page, and initials the citation list."
The better lines are longer. They're also testable.
A supervisor can look at a file and see whether the rule was followed. That's what turns a policy into supervision.
Who Should Own the Firm's AI Policy?
A named attorney with authority over the people using the tools should own it. In a small PI firm, that's usually the managing attorney. The owner approves tools, answers questions, runs training and sets the review date. A policy without an owner drifts. Nobody updates it when a vendor changes its terms.
The owner doesn't have to be technical. They do need to read the vendor terms, or have someone qualified read them. California's guidance says reasonable efforts may include consulting IT or cybersecurity experts.
How Do You Roll Out an AI Policy?
Start with what people already use. Then approve tools, write one page, train everyone who touches a tool, and set a review date. The order matters. A policy written before you know current habits will ban things nobody does and miss what they do.
- Ask what's in use. Every person lists the AI tools and accounts they've used for work, including free ones.
- Review the tools. Read each vendor's terms on training, retention and access. Mark which are approved for client data.
- Write one page. Use the eight sections above. Put the tool list and the data table in appendices.
- Train before use. Walk through the data table and the verification rule with real examples. Keep a sign-in sheet.
- Check a file. A month in, pull a few AI-assisted files and see whether the verification step happened.
- Set the review date. Put it on the calendar. Revisit it sooner when you add a tool or a vendor changes its terms.
When This Advice Is Wrong
I've written this for a small or mid-size PI firm. Here's where it may not fit.
- A solo practice. If you're the only person using AI, eight sections may be more than you need. Keep the data table and the verification rule. Those protect your clients whoever is at the keyboard.
- Client rules that go further. Some clients set their own AI limits. California's guidance says you must follow client instructions that restrict AI use. Theirs override yours.
- Changing guidance. I've quoted California, Washington and the NYC Bar. Your state may say more, and all of it keeps moving. When it changes, your policy should too.
What We Promise Instead
We don't claim our AI makes no mistakes. That's exactly why your policy needs a verification rule, and why we build for it. Our position is narrow: AI for assembly, humans for verification and strategy.
In practice, that means a draft your reviewer can check without rereading the file. Each fact points back to the record it came from.
So the verification line in your policy becomes a step someone can actually finish. Our post on Fact-Lock explains how that works and why we won't claim zero.
Where Lexyno Fits
Lexyno doesn't value a case, decide causation or give advice. Those stay with the attorney, and your policy should say so.
Lexyno's AI demand letter software drafts the demand from the case's own records, with source references your reviewer can check. Medical Summary builds the treatment chronology and flags gaps over 30 days. Plans are on the pricing page.
If you're deciding what goes on your approved list, our guide to AI for personal injury lawyers covers what each kind of tool is good for, and our medical chronology guide shows what a reviewer checks. For your team's side of the change, read our post on paralegals and AI. The bigger question is in Will AI Replace Lawyers?
One Page, One Owner, One Date
AI policies for law firms don't need to be long. They need an owner, a data table and a verification rule people actually follow.
Write the page this week. The next question at your door should already have an answer.
See how a demand is built in our demand letter sections guide, start from our PI demand letter template, or try Lexyno on one case from your own file, free.
Frequently asked questions
If anyone at the firm uses AI tools, yes in practice. California’s 2026 bar guidance says managerial and supervisory attorneys should establish clear policies on generative AI and train staff. Washington’s bar calls an AI use policy handbook and regular training best practice.
An owner, a list of approved tools and accounts, rules for what client data may go into each tool, verification rules for citations and numbers, attorney sign-off, client communication, billing, and a training and review schedule. A short policy people follow beats a long one nobody reads.
Only tools the firm has approved after checking their terms and security against your confidentiality duty. California’s 2026 guidance says an attorney must not input confidential client information into an AI tool that presents material risks to confidentiality or security, absent informed client consent.
No. In a Washington bar hypothetical, the firm’s IT team chose a tool with confidentiality protections, but the partner rolled it out with a brief email and no real training. The motion that followed cited misquoted cases. A policy covers how people use the tool, not just which tool.
It depends on your state and the use. California’s 2026 guidance says an attorney must consider disclosing the intent to use AI in the representation, and must follow any client instructions that limit AI use. Build both into the policy and your engagement letter process.
Set a date and keep it. California’s guidance says supervising attorneys should review and update firm policies, controls and training as AI technology evolves. Review it at least when you add a tool, when a vendor changes its terms, or when your bar issues new guidance.
Related posts

Legal Document Management Software: Security for PI Records
How to choose legal document management software for a PI firm: the security checks that matter for medical records, and where records leak instead.

Personal Injury Case Management Software: How to Choose
How to choose personal injury case management software: test deadlines, records, liens and settlement money on a closed file before you sign.

ChatGPT for Lawyers: What PI Firms Can Safely Use It For
ChatGPT for lawyers at a PI firm: what to paste and what not to, what the bar says about client data, and how to check a citation before you file it.
What's next?
Ready to see the difference in your next case?
Put these strategies to work. Try a mock demand generated by Lexyno today.
Request Evaluation Access