Client Financials and AI: Data Rules for a Fractional CFO Firm

Before anyone on your team pastes a client's P&L into an AI tool, your firm needs a few written rules. Client financials are confidential, many fractional CFOs are certified public accountants (CPAs) bound by professional rules, and consumer AI accounts are not the same as business ones. This post sets out what the rules say, what the main AI vendors state about your data, and a one-page policy you can adapt.

This is practical guidance, not legal advice. Check your engagement letters and, if needed, your own counsel.

What the professional rules say

If you or your CFOs are members of the American Institute of CPAs (AICPA), the Code of Professional Conduct's Confidential Client Information Rule (1.700.001) applies. A Journal of Accountancy article (Mar 2015) explains that confidential client information is any information obtained from the client that is not available to the public, and that a member in public practice shall not disclose it without the client's specific consent.

The same article covers Interpretation 1.700.040 on third-party service providers: providers you use to deliver work, which an AI vendor arguably is. Because clients may not expect you to use one, you must either:

• have a contract with the provider that requires confidentiality and gives you reasonable assurance it has procedures to prevent unauthorized release, or
• get the client's specific consent before disclosing their information to the provider.

The article is archived and notes its content may be out of date, so check the current Code on the AICPA site. If your firm also prepares tax returns, the article notes that Internal Revenue Code Section 7216 rules on tax return information apply separately.

If your CFOs are not CPAs, these rules may not bind you directly, but your clients will still expect the same care, and your engagement letter may promise it.

What the AI vendors say about your data

The single biggest difference is between consumer accounts and business accounts.

• OpenAI states for its business products: "By default, we do not use your business data for training our models" (OpenAI enterprise privacy, updated Jan 2026). Data you share through explicit opt-in mechanisms, such as feedback, may be used.
• Anthropic states that by default it will not use inputs or outputs from its commercial products, such as Claude for Work and the API, for training, unless you submit feedback or otherwise opt in (Anthropic privacy center, Aug 2026).

Two practical consequences. First, a personal or free chat account is not where client financials belong. Second, the thumbs up and thumbs down buttons can count as feedback, so tell staff not to use them on client work.

Training is only one question. Ask each vendor about data retention, where data is stored, who at the vendor can access it, and whether they will sign a confidentiality or data processing agreement. That last one is what lets you rely on the contract route in 1.700.040 rather than asking every client for consent.

Your reporting tools count too

The same thinking applies to reporting and forecasting tools that have added AI features. If your reporting tool sends data to an AI model to write commentary, read its terms the same way. Fathom says its commentary feature traces each number back to its source (Fathom); that helps accuracy but doesn't answer where the data goes. Ask.

A one-page AI policy for a fractional CFO firm

Adapt this to your firm. Keep it to one page so people actually read it.

1. Approved tools. List each AI tool and account type staff may use with client data (for example, "Claude Team account under the firm's name" or "Fathom Pro Commentary Writer"). Anything not listed is not approved for client data.

2. Never in AI. Personal or free accounts of any AI tool. Bank login details and passwords. Tax return information unless your Section 7216 process covers it. Personal information about client employees beyond what the work needs.

3. Client consent or contract. For each approved tool, note whether you rely on a vendor confidentiality agreement or on client consent. If consent, record which clients agreed and when. Add a sentence to new engagement letters naming the categories of tools you use.

4. Figures come from data. AI drafts text. Figures are pulled from the ledger, reporting tool or a spreadsheet, not calculated by the model.

5. A CFO reviews everything. No AI-drafted text reaches a client, board or lender without review by the responsible CFO.

6. No feedback buttons on client work. Unless your vendor agreement covers it.

7. Who approves new tools. Name one person. Staff ask before trying a new AI tool on client data.

8. Review the policy. Set a date, at least once a year and whenever a vendor's terms change.

How to roll it out

1. Inventory what's in use now. Ask staff, without blame, which AI tools they already use and for what. You'll likely find personal accounts.
2. Pick one business account and move work there.
3. Update engagement letters at renewal.
4. Brief the team in 30 minutes and put the policy where they'll see it.
5. Check quarterly for new tools creeping in.

Why this helps you sell, too

Clients are starting to ask how their advisors use AI. A CFO quoted in CFO Dive (Aug 2026) described helping clients adopt AI as "a new responsibility for the CFO." A firm that can hand over a clear one-page policy answers the question before it becomes a worry, and has a ready template for clients who want their own.

With the rules in place, you're ready to try the board-pack narrative template. For the wider picture, see how fractional CFOs use AI.

---

Want a second pair of eyes on your monthly cycle? Lunan offers a free conversation: we ask how your close-to-board-pack month runs, which tools you use and how you bill, then send a short written note on where drafting would save the most time and where your current tools already cover it. Start a free conversation.