Key takeaways
- ABA Formal Opinion 512 says managerial lawyers must establish clear policies on permissible generative AI use and ensure training[1].
- The training gap is wide: 64% of professionals had no training on generative AI for their work[2].
- A workable policy fits on a few pages: approved tools, data rules, verification, client communication, billing, supervision and incidents.
- Reflect your state's guidance; California now addresses agentic AI, including autonomous client intake[3][4].
Why your firm needs a written policy
Staff are already using generative AI, whether or not the firm has approved it. A written policy turns individual habits into firm practice. It also meets a professional duty: ABA Formal Opinion 512 says managerial lawyers must establish clear policies on the firm's permissible use of generative AI, and supervisory lawyers must make reasonable efforts to ensure lawyers and nonlawyers comply, including training in the ethical and practical use of the tools and their risks[1].
A policy does not need to be long. A small firm can cover the essentials in three or four pages. The test of a good policy is whether a new paralegal could read it on day one and know which tool to use, what they may put into it, and who checks the result.
The training gap
Thomson Reuters found that although 89% of professionals see use cases for generative AI, 64% have received no training on using it for their work[2]. A policy without training is a document nobody follows. Plan the training session at the same time as you write the policy.
Source: Thomson Reuters, 2025[2]
Show the data
| Item | Value |
|---|---|
| See use cases for it | 89% |
| Have had no training on it | 64% |
The sections every policy needs
Purpose and scope
Say who the policy covers (lawyers, staff, contractors) and what counts as AI, including features built into software the firm already uses.
Approved tools
Refer to a register of approved tools and approved uses. Anything not on it needs approval before use.
Data rules
State what information may be entered into which tools, using simple categories.
Verification
Require human review of all output before it is relied on, with stricter checks for research, citations and anything filed or sent to a client.
Client communication and consent
Explain when clients are told about AI use and when informed consent is needed, and who decides.
Billing
Bill for time actually spent; do not charge clients for learning a tool the firm uses routinely.
Supervision and training
Name who supervises AI use, what training is required and how often it is refreshed.
Incidents
Explain how to report a suspected data exposure or an error that reached a client or court.
Review
Set a review date, at least yearly and whenever your state issues new guidance.
An approved-tools register
| Tool (Example) | Approved uses | Data allowed | Owner | Next review |
|---|---|---|---|---|
| Enterprise drafting assistant | Plain-language rewrites, first drafts from firm templates | Client data allowed under enterprise terms | Managing partner | Every six months |
| Transcript summary feature in practice software | Summaries of documents already in the file | Client data allowed | Litigation lead | Every six months |
| Consumer chatbot accounts | Marketing outlines, general research leads | No client or matter information | Office manager | Yearly |
| AI phone answering | Answering intake calls with firm-approved questions | Caller details collected under the intake script | Intake lead | Quarterly |
Keeping the register separate from the policy means you can add or retire a tool without rewriting the whole document. Record the date each tool's terms were last checked, because vendors change retention and training terms without much notice.
Data categories staff can apply quickly
- Public: published law, your own website content, general questions. Any approved tool.
- Internal: firm procedures, templates, non-client business information. Approved business or enterprise tools only.
- Client confidential: anything relating to a representation, including names and facts. Only tools approved for client data, and only where the policy allows without consent or consent has been obtained.
- Restricted: health records, financial account numbers, sealed material, information under protective order. Not entered into any AI tool without written approval from the responsible lawyer.
California's guidance says lawyers must not input confidential client information into a tool that may present material risks to confidentiality or security without informed client consent[5]. Your categories are how staff apply that rule in a hurry.
Sample wording
Sample wording (adapt before use): "Staff may use only tools listed in the AI Tools Register, for the uses listed. Before relying on any AI output, the user must check it against the source material. Any legal authority suggested by an AI tool must be read in full by a lawyer before it is cited."
Sample wording (adapt before use): "No information relating to a client matter may be entered into a tool marked 'No client data' in the Register. If you are unsure, ask the responsible lawyer before entering anything."
Sample wording (adapt before use): "Time recorded for AI-assisted work must reflect the time actually spent. Time spent learning a tool the firm uses routinely is not billed to clients."
State guidance to reflect
- California: the State Bar's 2026 Practical Guidance replaces its November 2023 version and, at the California Supreme Court's request, addresses agentic AI[3]. It names autonomously facilitating client intake as an example and says supervision must correspond to the system's level of access and autonomy[4].
- Florida: Ethics Opinion 24-1 says lawyers may use generative AI but must protect confidentiality, provide accurate and competent services, avoid improper billing and comply with advertising rules[6]. Client-facing chatbots must disclose that they are an AI program and not a lawyer or employee of the firm[7].
- Texas: Professional Ethics Committee Opinion 705 addresses lawyers' use of generative AI under the Texas Disciplinary Rules[8].
- Your state: check your bar's ethics opinions and add a line for each that applies.
Approving a new tool
- The requester describes the task, the data involved and the expected benefit.
- The owner reads the terms of use and privacy policy, or asks someone qualified to, and records retention, training use and access by the vendor.
- The responsible lawyer decides the data category allowed and whether client consent is needed.
- The tool is tested on non-client or closed-matter material, and errors are logged.
- If approved, it is added to the register with uses, data allowed, owner and review date, and staff are told.
Training that sticks
Keep training short and practical: 45 minutes, built around the firm's actual tools. Cover what the tools are good and bad at, a live example of a hallucinated citation, the data categories, the verification steps and how to report a problem. Repeat it when a new tool is approved and once a year for everyone. Keep a simple attendance record, which shows supervisors met their duty if a question ever arises.
Example: a six-lawyer general practice
Example (fictional): a six-lawyer firm finds that three staff use personal chatbot accounts for drafting emails, sometimes with client names. The managing partner writes a four-page policy and a register with three approved tools, moves drafting to an enterprise account whose terms exclude training on inputs, and runs a 45-minute session. Two months later, a paralegal reports that a summary tool mixed up two clients' dates. The incident process works as written: the summary is corrected before it is sent, and the policy adds a rule against summarizing more than one matter in a session.
Common policy mistakes
- Writing a policy so broad that nobody can tell whether a given use is allowed.
- Approving a tool without recording which plan the firm is on, when retention and training terms differ by plan.
- Covering lawyers but not staff, contractors or interns.
- Ignoring AI features inside existing software, such as transcription in a phone system or summaries in practice management.
- No incident process, so mistakes are fixed quietly and nothing is learned.
- Treating the policy as finished once it is signed.
The fix for most of these is the same: keep the policy short, keep the register specific, and name an owner who updates both. Ask one staff member to read a draft and say which uses they still cannot classify, then rewrite those parts.
Keep it current
Review the policy at least once a year and whenever your state bar issues new guidance, a vendor changes its terms, or an incident shows a gap. Date every version, keep the old ones, and send staff a short note explaining what changed and why. A policy that is never updated becomes a policy that nobody reads.
Frequently asked questions
Does a law firm need an AI policy?
ABA Formal Opinion 512 says managerial lawyers must establish clear policies regarding the firm's permissible use of generative AI[1]. Even a short written policy is better than none.
What should an AI use policy include?
Scope, approved tools, data rules, verification, client communication and consent, billing, supervision and training, incident reporting and a review date.
Should the policy ban AI entirely?
A ban is hard to enforce and usually pushes use out of sight. Approving specific tools for specific tasks gives the firm more control.
How often should the policy be reviewed?
At least yearly, and whenever state guidance changes or an incident reveals a gap.
References
American Bar Association, Standing Committee on Ethics and Professional Responsibility (2024). Formal Opinion 512: Generative Artificial Intelligence Tools (archived copy) (opens in a new tab) โ
Thomson Reuters (2025). Generative AI Adoption Nearly Doubles as Professional Services Reach Crossroads (opens in a new tab) โ
The State Bar of California, Standing Committee on Professional Responsibility and Conduct (2026). Practical Guidance for the Use of Generative Artificial Intelligence in the Practice of Law (opens in a new tab) โ
The State Bar of California, Standing Committee on Professional Responsibility and Conduct (2026). Practical Guidance for the Use of Generative Artificial Intelligence in the Practice of Law (opens in a new tab) โ
The State Bar of California, Standing Committee on Professional Responsibility and Conduct (2026). Practical Guidance for the Use of Generative Artificial Intelligence in the Practice of Law (opens in a new tab) โ
The Florida Bar (2024). Florida Bar Ethics Opinion 24-1 (opens in a new tab) โ
The Florida Bar (2024). Florida Bar Ethics Opinion 24-1 (opens in a new tab) โ
Texas Center for Legal Ethics (Professional Ethics Committee for the State Bar of Texas) (2025). Opinion 705 (opens in a new tab) โ