AI policy for tax firms: how to add AI to your WISP
Most small tax, accounting and bookkeeping firms already use AI somewhere: drafting client emails, summarizing IRS notices, cleaning up bank feeds. Very few have written down the rules. Since the IRS Office of Professional Responsibility published its first AI guidance in June 2026, that gap has become hard to ignore.
Why tax firms need written AI rules now
On June 24, 2026 the IRS Office of Professional Responsibility (OPR) issued Alert 2026-19, "Introductory Guidelines for Responsible AI Use in Federal Tax Practice". It does not create new rules. Instead, it explains how duties practitioners already have under Treasury Circular 230 apply when AI is involved:
- Due diligence (§10.22): you must check AI-produced work, including facts, citations and calculations, before it reaches a client or the IRS. Relying on the tool alone isn't enough.
- Competence (§10.35): you need to understand how the tools you use produce their answers, and where they get things wrong.
- Firm procedures (§10.36): the people in charge of a firm must put reasonable procedures in place. For AI the alert points to staff training, secure data-handling rules, vetting of third-party AI tools, and documentation that shows the steps were followed.
- Written advice (§10.37): advice can't rest on unverified AI output.
- Fees (§10.27): billing for time not actually spent because AI did the work can be a problem, depending on the facts.
- Confidentiality: unauthorized use or disclosure of tax return information can bring civil and criminal penalties under IRC §§6713 and 7216, and client data should only go into secure, firm-approved AI systems.
Section 10.36 is the one that turns this into paperwork. If a firm is ever asked how it manages AI risk, "we told people to be careful" won't hold up. A written policy, a training record and an AI use log will.
Why the AI policy belongs in your WISP
Under the FTC Safeguards Rule (16 CFR Part 314), professional tax preparers count as financial institutions, so they must keep a written information security plan. IRS Publication 5708 gives small practices a sample WISP, and IRS Publication 4557 explains the security basics.
Your WISP already covers the things AI touches: who is responsible for security, how you assess risk, how you oversee service providers, how you train staff and how you respond to incidents. An AI tool is, in practice, another service provider handling client data. So the cleanest approach for a small firm is an AI addendum to the WISP, not a free-standing policy nobody reads.
The 10 sections every tax-firm AI policy needs
1. Purpose, scope and owner
Say what the policy covers (any AI tool used for firm work, including AI features built into software you already use), who it applies to (staff, contractors, seasonal preparers) and who owns it. For most firms the owner is the WISP's Qualified Individual.
2. Approved AI tools register
Keep a short list of approved tools and the exact plan or tier you pay for. Business and consumer tiers of the same product often have different data terms. Anything not on the list is not approved for client work.
3. Data rules: what can go into AI
A simple traffic-light scheme works well:
| Level | Examples | Rule |
|---|---|---|
| Green | Public information, generic templates, marketing drafts with no client details | Any approved tool |
| Amber | Client facts with names, SSNs and account numbers removed | Approved tools only, after redaction |
| Red | Returns, SSNs/ITINs, bank details, identity documents, anything that identifies the client | Only tools specifically approved for client data, and only where the client has consented |
4. Vendor vetting
Before approving a tool, find out: does the provider keep your prompts and files, for how long, and does it train its models on them? Where is data stored? Is one client's data kept separate from another's? Can you get these answers in writing for the plan you actually use? Keep the answers in a vendor file. If a provider won't confirm in writing that it doesn't train on your data, treat that as a no.
5. Client consent and a no-AI option
Section 7216 generally requires a client's written consent before a preparer uses or discloses tax return information outside the exceptions in the regulations. The IRS hasn't said whether sending return information to an AI tool counts as a disclosure to a third party. The cautious path many firms are taking is a standalone, signed AI consent drafted with their attorney. Consent has to be voluntary, so you also need a way to serve clients who say no, and a flag so staff know who they are.
6. Human review
Treat all AI output as a first draft. Name who reviews what: a preparer checks figures against source documents, a senior reviewer checks any research or citations, and nothing AI-drafted goes to a client or the IRS unreviewed.
7. AI use log
For each engagement, record which AI tools touched the client's information, what for, and who reviewed the result. This is the documentation the OPR alert asks firms to keep, and it also tells you how much time AI is really saving.
8. Billing
Bill for the work actually done. If AI cuts a two-hour task to twenty minutes, don't bill two hours of manual time. Say how your firm handles this, including whether you tell clients about AI-assisted work.
9. Training and sign-off
Train every person before they use AI for firm work and again each year, including seasonal staff. Keep a signed acknowledgement, because the training record is your evidence under §10.36.
10. Incidents and annual review
Define an AI incident (client data in an unapproved tool, an AI error that reached a client or the IRS) and plug it into the incident-response steps your WISP already has. Note that the Safeguards Rule now requires notifying the FTC when certain security events affect 500 or more consumers. Review the AI policy at least once a year and whenever you add a tool.
How an AI policy maps to the FTC Safeguards Rule
| Safeguards Rule element (16 CFR 314.4) | Where AI fits |
|---|---|
| Qualified Individual | Owns the AI policy, tools register and vendor file |
| Written risk assessment | Add AI risks: data leaving the firm, retention, model training, errors |
| Safeguards (access controls, MFA, encryption) | Approved tools only, business accounts with MFA, no personal AI accounts for client work |
| Testing and monitoring | Spot-check the AI use log and outputs |
| Staff training | AI rules training with sign-off, including seasonal staff |
| Service-provider oversight | Vendor vetting questions and written data terms for each AI tool |
| Keeping the program current | Annual AI policy review and review when tools or guidance change |
| Incident response plan | AI incidents defined and routed into the existing plan |
The table is a planning aid, not a legal analysis. Read the rule itself and Publication 5708 for the full requirements.
Skip the blank page
The PracticeSafe AI Policy Add-On is a ready-to-edit AI addendum for your WISP: an AI acceptable use policy with these 10 sections, a vendor review checklist with 18 questions, a staff training guide with quiz and sign-off, and client consent templates including a no-AI letter.
See the AI Policy Add-On ($79)Get the free 15-point checklistA 7-step rollout plan for a small firm
- Find out what's already in use. Ask everyone which AI tools they use for work, including AI features inside your tax, practice-management and email software.
- Pick your approved tools and check the data terms for the plan you'll actually pay for.
- Write the AI addendum using the 10 sections above, and attach it to your WISP.
- Decide your consent approach with your attorney, and set up the no-AI path.
- Start the AI use log: even a shared spreadsheet works.
- Train the team and collect signed acknowledgements before tax season.
- Put the annual review in the calendar, alongside your WISP review.
Not sure where your firm stands? The free AI policy generator builds a starter policy from a few questions, and the free AI Safety Checklist gives you 15 quick checks.
Frequently asked questions
Do tax preparers legally need an AI policy?
No law uses the words "AI policy". But the FTC Safeguards Rule already requires a written information security plan (WISP) covering how client data is protected, and IRS OPR Alert 2026-19 says firm leaders must have procedures, training and documentation for AI under Circular 230 section 10.36. Written AI rules are the practical way to show both.
Should the AI policy be a separate document or part of the WISP?
Most small firms are best served by an AI addendum to the existing WISP. The WISP already names your Qualified Individual, risk assessment, vendor oversight and incident response, and AI use touches every one of those.
Can my staff use free ChatGPT for client work?
Not with client data. The IRS alert says client information should only go into secure, firm-approved AI tools, and consumer AI tiers often keep or train on what you type. Free tools can still be fine for work with no client data, such as drafting a generic newsletter.
Do I need client consent to use AI?
Possibly. Section 7216 generally requires a client's written consent before a preparer uses or discloses return information outside the regulatory exceptions, and the IRS has not said whether sending data to an AI tool is a third-party disclosure. Many firms are choosing a standalone, signed AI consent drafted with their counsel, plus a no-AI option for clients who decline.
How often should we review our AI policy?
At least once a year, and whenever you add a new AI tool or the IRS, FTC or your state issues new guidance.
Sources
- IRS OPR Alert 2026-19: Introductory Guidelines for Responsible AI Use in Federal Tax Practice (June 24, 2026)
- FTC Safeguards Rule, 16 CFR Part 314
- IRS Publication 5708, Creating a Written Information Security Plan for your Tax & Accounting Practice
- IRS Publication 4557, Safeguarding Taxpayer Data
- Treas. Reg. 301.7216-3 (consent to disclose or use tax return information)