AI Acceptable Use Policy Template for Law Firms and RIAs
A copyable AI acceptable use policy for firms that handle client confidences and regulated records, with notes for law firms (ABA Opinion 512) and RIAs and broker-dealers (FINRA 24-09), and the clause most policies leave out: knowing where AI is actually running.
An AI acceptable use policy tells everyone at the firm which AI tools they may use, with what data, under what review, and who is accountable for the result. For a law firm or a wealth management firm it is also the written record that professional duties, client confidentiality and supervision rules have been turned into practice. The template below is written for both. Replace the bracketed placeholders, cut what does not apply, and have it reviewed by your general counsel or compliance team.
This template is a starting point, not legal advice. Your obligations depend on your jurisdiction, registrations and client agreements.
Why firms need one now
AI use at most firms started bottom-up: someone tried an assistant, it helped, and it spread. By the time leadership asks what the policy is, people are already using several tools on real work, and some teams have started building agents and automations of their own. A policy written at that point has two jobs: set the rules going forward, and find out what is already running.
| Risk | What goes wrong without a policy | Clause that addresses it |
|---|---|---|
| Confidentiality | Client information pasted into a personal account with unknown terms | 3, 4 |
| Accuracy | An invented citation or figure reaches a client or a filing | 5 |
| Disclosure | A client learns after the fact that AI was used on their matter | 6 |
| Records | No record of what an AI tool produced or decided | 7, 10 |
| Shadow agents | Automations touching client systems with no owner | 9 |
The template
Copy the clauses below into your firm's policy format. Placeholders are in brackets.
1. Purpose and scope
1.1 This policy governs the use of artificial intelligence tools, including generative AI assistants, AI agents and AI-driven automations, by all partners, employees, contractors and temporary staff of [Firm Name] ("the Firm") when performing Firm work or handling Firm or client information.
1.2 The policy applies on Firm devices and personal devices, and whether a tool is provided by the Firm or accessed independently.
1.3 [Policy Owner, e.g. Chief Operating Officer / General Counsel / Chief Compliance Officer] owns this policy and approves exceptions in writing.
2. Definitions
2.1 "AI tool" means any software that generates text, images, code, analysis or decisions using a machine learning model.
2.2 "AI agent or automation" means any AI tool configured to act on Firm systems, such as reading or updating records, filing documents or sending communications, with or without a person present for each action.
2.3 "Confidential information" means client information, matter or account information, personal information, and any Firm information not approved for public release.
3. Approved tools and plans
3.1 Only the AI tools and plans listed in Schedule A may be used for Firm work. Schedule A is maintained by [Policy Owner] and lists, for each tool, the approved plan (for example, a business or enterprise tier), the permitted data classifications and the permitted uses.
3.2 Personal or consumer accounts of any AI tool must not be used for Firm work or with any Firm or client information.
3.3 Requests to add a tool are submitted to [Policy Owner / IT Security] and reviewed for vendor terms, data retention, training on customer data, security attestations and admin controls before approval.
4. Data you may and may not enter
4.1 Public information and the Firm's own non-confidential material may be used with any approved tool.
4.2 Confidential information may be entered only into tools approved for it in Schedule A, and only to the minimum extent needed for the task.
4.3 The following must never be entered into any AI tool unless Schedule A expressly permits it: [e.g. government identification numbers, account credentials and passwords, health information, information subject to a protective order or a client restriction].
4.4 Where a client agreement or outside counsel guideline restricts AI use, that restriction overrides this policy for that client.
5. Human review
5.1 AI output is a draft. The person using the tool is responsible for it as if they had written it themselves.
5.2 Every citation, quotation, figure, date and statement of fact in AI output must be checked against a primary source before it is relied on.
5.3 No AI output may be sent to a client, filed with a court or regulator, or used in a client recommendation without review by [a qualified person / the responsible attorney or adviser].
6. Client disclosure and consent
6.1 The Firm discloses its use of AI to clients as described in [engagement letter / client agreement / Form ADV / privacy notice].
6.2 Before confidential client information is entered into a generative AI tool, [the responsible attorney / adviser] confirms whether client consent is required and obtains it where it is.
6.3 AI use does not change how time is recorded or fees are charged except as approved by [Billing Committee / Management].
7. Recordkeeping
7.1 AI output that becomes a Firm record, client communication or the basis of a decision is retained under the Firm's records policy, like any other record.
7.2 Retention settings on approved AI tools are configured by [IT / Compliance] to match the records policy.
8. Approved and prohibited uses
8.1 Approved uses include: [drafting and editing internal documents; summarising documents the user is permitted to access; research support subject to clause 5; meeting notes where all participants have been informed; operational automations registered under clause 9].
8.2 Prohibited uses include: [making final legal, investment or credit decisions without human review; generating client-facing advice without review; impersonating a person; circumventing access controls; any use that breaches law, regulation or client terms].
9. Building agents and automations
9.1 Only [approved builders / teams listed in Schedule B] may build AI agents or automations that act on Firm systems.
9.2 Every agent or automation is registered before it runs on Firm or client data, with: a named owner, the team it serves, its purpose, the systems and permissions it uses, and the platform it runs on.
9.3 An agent without a current owner is paused until one is assigned.
9.4 Agents use the narrowest access that does the job, under the permissions of the person or service account approved for it.
10. Monitoring and audit
10.1 The Firm monitors use of approved AI tools through their admin consoles, and the runs of registered agents and automations through their run records.
10.2 For any agent or automation that touches confidential information, the Firm must be able to trace what it did on a given run: what it received, which systems it touched, under which permission, and what it changed.
10.3 [Compliance / Internal Audit] reviews the register in clause 9.2 and a sample of runs [quarterly].
11. Training
11.1 Everyone covered by this policy completes AI training before using an approved tool on Firm work, and [annually] after that.
12. Incidents
12.1 Report suspected misuse, data exposure or materially wrong AI output that reached a client, court or regulator to [Incident Contact] within [24 hours].
13. Review
13.1 This policy and its schedules are reviewed [every six months] and whenever the Firm approves a new category of AI tool.
Schedule A: Approved tools, plans, data classifications and uses.
Schedule B: Approved builders and the agent register.
Notes for law firms
The ABA's Formal Opinion 512 (July 2024) is the anchor for most law firm AI policies. It says lawyers using generative AI must understand the tool's capabilities and limits (competence), protect client information (confidentiality), communicate with clients about its use, and charge reasonable fees. It recommends informed client consent before client confidences go into a generative AI tool, and says boilerplate consent in an engagement letter is not enough. That is why clause 6 asks the responsible lawyer to confirm consent rather than relying on the engagement letter alone.
State bars have issued their own guidance and it differs, so check your jurisdiction. Clause 5 matters more in law than anywhere else: courts have sanctioned lawyers for filing citations an AI tool invented. Our guide to AI hallucinations in law covers why it happens and how firms check for it.
Notes for RIAs and broker-dealers
FINRA's Regulatory Notice 24-09 (June 27, 2024) says its rules, which are intended to be technology neutral, continue to apply when member firms use generative AI. The content standards for communications under Rule 2210 apply whether a communication is written by a person or a tool, and a firm using generative AI in its supervisory system should address technology governance, including model risk management, data privacy and integrity, and reliability and accuracy, under Rule 3110.
Recordkeeping for AI prompts and outputs is the open question. Regulators have not squarely resolved it, and existing books-and-records and supervision obligations still apply. Clause 7 takes the conservative path: anything that becomes a record or a client communication is kept like one. Registered investment advisers should also reconcile the policy with their compliance manual and Form ADV disclosures.
The clause most policies leave out: knowing where AI runs
Most AI policies are written about assistants: what you may type into a chat window. The harder problem arrives when teams start building agents and automations that act on client systems. A policy can say every agent needs an owner, but it can only be enforced if the firm can see the agents. That is clause 9, and it is the one that tends to exist on paper only.
Two questions decide whether it works. Where is AI doing work today, and where should it? And for each agent that runs, who owns it, what did it do, and can compliance trace it?
Where Caddi fits
- Caddi Discover maps the work across the firm's systems and recommends where each task should run: in Caddi as deterministic code for high-volume, high-accuracy work, or with Claude or Copilot for individual, judgment-heavy work.
- Caddi Govern keeps one record of every Caddi agent and the work Caddi hands to Claude as a skill or to Copilot as an agent, with owner, team and an audit log compliance can trace decision by decision. That is the register clause 9 asks for and the run record clause 10 asks for, for the work that runs through Caddi.
Govern does not see a person's ad hoc chats, or agents built in Claude, Copilot or other platforms without Caddi. Those stay under clause 3 and each vendor's admin console. If you are choosing the assistant to approve in Schedule A, see ChatGPT vs Claude vs Copilot for lawyers or for financial advisors.
Caddi
See how Caddi AI Agents can find your most repetitive back-office work and automate it across every tool you use
Frequently asked questions
What should an AI acceptable use policy include?
At minimum: scope, the approved tools and plans, what data may and may not be entered, human review requirements, client disclosure and consent, recordkeeping, approved and prohibited use cases, rules for building agents and automations, monitoring and audit, training, incident reporting and a review cadence. The template on this page covers each one with bracketed placeholders to fill in.
Does a law firm need a generative AI policy?
Practically, yes. ABA Formal Opinion 512 (July 2024) says lawyers using generative AI must understand the tool, protect client information, communicate with clients about its use and charge reasonable fees. A written policy is how a firm shows it has turned those duties into rules its people follow. State bars may add their own guidance.
What does FINRA say about generative AI?
FINRA Regulatory Notice 24-09 (June 27, 2024) says its rules are technology neutral and continue to apply when member firms use generative AI. Communications standards under Rule 2210 apply whether content is written by a person or a tool, and firms using generative AI in supervision should address technology governance, including model risk management, data privacy and accuracy.
Do AI prompts and outputs have to be kept as records?
Regulators have not squarely resolved it. Existing books-and-records and supervision obligations still apply, so many firms treat AI output that becomes a business communication or a record of a decision as they would any other record, and configure retention on their business AI plans accordingly. Ask your compliance counsel how your obligations apply.
Should employees be allowed to use personal ChatGPT or Claude accounts?
Not for firm or client information. Business tiers such as ChatGPT Business or Enterprise, Claude Team or Enterprise and Microsoft 365 Copilot come with admin controls and vendor terms the firm can review. A personal account gives the firm neither. Most policies approve specific business plans and prohibit personal accounts for any firm work.
How do we know where AI is actually being used?
That is the hardest part of enforcing any AI policy. A registration rule helps: every agent or automation that touches firm systems is registered with an owner, a purpose and where it runs. Caddi Discover maps the work across the firm and recommends where each task should run, and Caddi Govern keeps one record of every Caddi agent and the work Caddi hands to Claude and Copilot, with owner, team and audit log.