Can an Accounting Firm License Its Data to AI Companies?

Can an accounting firm license its data to AI companies?

An accounting firm can license its own work product and should expect to decline on client information. That split is not a cautious reading of the rules, it is the reading the AICPA's own ethics staff published this month.

Writing in the Journal of Accountancy on November 2, 2026, Kelly D. Mullins, a communications manager for the AICPA Professional Ethics Division, addresses offers to use firm data for AI training directly and says they are rising. The recommended default is "to decline unless you can clearly establish that the transaction is permissible". The guidance adds that "A sale or other transfer intended to train an AI model does not eliminate that specific-consent requirement", and that labelling the buyer a technology partner does not resolve the question.

The rule underneath it is the Confidential Client Information Rule, ET sec. 1.700.001, which the same guidance says generally prohibits members in public practice from disclosing confidential client information without the client's specific consent. ET sec. 1.700.060, on disclosure of client information to third parties, "applies even when client information is disclosed without the client being specifically identified".

So the answer has two halves and most firms only hear the first. No, a firm should not expect to license client files, workpapers or returns, and de-identifying them does not change that. Yes, a firm can license what it made itself, which is the part of the archive that shows how this firm thinks. The rest of this page is about telling the two apart and what the second is worth.

Why is redaction not enough for an accounting firm?

Redaction is not enough because confidentiality under the AICPA Code attaches to the information rather than to the name on it, and the Professional Ethics Division said so in terms on November 2, 2026: "Redaction does not, by itself, turn confidential client information into nonconfidential information."

That single sentence is why this industry is different from the rest of the data licensing market. In manufacturing or home services, de-identification is the step that makes a dataset licensable. The buyer strips names and account numbers and the deal proceeds. In public accounting, stripping the identifiers leaves confidential client information that is still confidential, and the consent requirement survives the redaction. The interpretation the guidance points to, ET sec. 1.700.060, says as much on its face: it applies even where the client is not specifically identified.

The same guidance warns against treating a buyer's "anonymous data" label as permission. Where information remains confidential, the client's specific consent is required, preferably in writing, and the guidance says any consent you seek "should clearly explain the information involved, the recipient, and the training purpose". It is not a line added to a technology disclosure in an engagement letter.

Two practical consequences follow. A buyer who answers your confidentiality question with a description of its anonymization pipeline has not answered it, and you should say so. And a firm that wants to pursue this route on client data is undertaking a client-by-client consent exercise, with the right to refuse sitting with each client, which is a far larger project than the sales process suggests.

The honest version of the pitch for an accounting firm is therefore narrower than the one being made to other industries. It is not "we anonymize it and you get paid". It is "there is a part of your archive this does not touch, and that is the part we can talk about".

What does Section 7216 add for a firm that prepares tax returns?

Section 7216 adds criminal exposure for a firm that prepares tax returns, on top of the ethics rules, and it applies to tax return information specifically. The AICPA guidance of November 2, 2026 states that "Sec. 7216 can impose criminal consequences" for knowing or reckless unauthorized disclosure or use of tax return information, advises having counsel evaluate the transaction, and warns against assuming "an ordinary engagement letter is sufficient".

The scope matters, because it is narrower than firms assume and the gap works both ways. Sec. 7216 reaches tax return information. Audit workpapers sit outside it and remain inside the AICPA Code, so a firm cannot clear an audit practice by establishing that the tax rules do not apply. A firm with both practices is answering two different questions about two different archives.

Written consent is the mechanism, and the Treasury regulations set conditions on it. Under 26 CFR 301.7216-3(a)(1), "The consent must be knowing and voluntary", and "conditioning the provision of any services on the taxpayer's furnishing consent will make the consent involuntary". A separate civil penalty under Sec. 6713 runs at $250 for each prohibited disclosure or use, capped at $10,000 in a calendar year, and unlike the criminal provision it does not require the conduct to be knowing or reckless. And 26 CFR 301.7216-2(d)(2) provides that "Contractors receiving tax return information pursuant to this section are tax return preparers under section 7216", which is directly relevant to any AI buyer receiving return data.

For most firms the practical effect is simple. Tax return information is out. Not because a consent route is theoretically unavailable, but because obtaining knowing, voluntary, specific written consent from every affected client for an AI training disclosure is a project with a worse ratio of effort to payment than any figure in this market justifies.

What can an accounting firm actually license?

An accounting firm can license what it created itself, and in a professional services business that is more than it sounds. The test is whether the material would exist without any particular client, and whether it can be described without conveying anything about one.

Four categories usually survive the analysis. The first is the methodologies and work programs the firm developed itself, and the second is the templates, checklists and standard workpaper structures built around them. Internal SOPs, training material and review standards are the third. The fourth is the firm's own operational records, meaning how work was scheduled, staffed, reviewed and resolved internally, with client identity and client information removed rather than merely masked.

That last category is where the market's interest sits, because it shows professional judgment being applied in sequence. micro1 lists SOPs, process documentation, internal knowledge bases, workflows, templates, CRM records and support tickets on its data partnerships page, read on November 10, 2026, names finance and accounting among its industries, and adds a category it calls agent trajectories: the record of how a task was actually carried out. Handshake's data partnerships page, read the same day, says "Your documentation, workflows, and project history capture how your business operates", and lists professional services among its industries.

The size of the opportunity should be set honestly against that. A firm licensing its own methodology and internal workflow is offering a smaller dataset than a manufacturer offering a decade of production records, and the number should be expected to reflect it. What makes it worth doing is that this part of the archive carries no consent problem, no Sec. 7216 question and no professional exposure, which makes it the only part worth a partner's time.

What has to stay out of an accounting firm's data license?

Everything a client gave you or that you produced about a client stays out unless that client has given specific consent for this purpose. For an accounting firm the exclusions are the deal rather than a schedule at the back of it, so here they are in full.

These stay out of any scope you describe to a buyer:

• Client financial records, general ledgers, trial balances and anything exported from a client's accounting system.
• Workpapers, including audit workpapers, which the AICPA Code reaches even where Sec. 7216 does not.
• Tax returns and tax return information of any kind.
• Client correspondence and advice given to a client.
• Client lists and engagement terms.
• Client material held under an NDA.

All of it remains confidential after redaction, per the November 2, 2026 guidance.

Also out on the standard grounds that apply across this lane: HR and payroll records for your own staff, health information, and call or video recordings made without all-party consent. Firms running client portals should check what the portal's own terms permit before an export is contemplated.

One category needs a partner's judgment rather than a rule. Anonymized aggregate statistics about the firm's practice, such as average cycle time on a return type, may be the firm's own operating data rather than client information. That is a line worth drawing with counsel on your actual facts, because the AICPA guidance's point is precisely that the test is not whether a name appears.

Practical sequence: run the exclusions first, describe only what is left to any buyer, and decline to discuss a scope that reaches client information even hypothetically. A buyer accustomed to other industries will ask. The answer does not change.

Who is buying accounting firm data, and what do they pay?

Programs aimed at accounting firms exist, and the one aimed squarely at the profession launched this autumn. Accounting Today reported on October 16, 2026, updated October 19, that BizBuyData, a program announced by Mode.Inc, lets firms license historical operational records to AI labs training agents to perform accounting and business tasks. "Firms with at least 10 employees using digital workplace tools are eligible." Payment "varies based on the amount and type of eligible data" and is expected within approximately 60 days, and Mode collects and de-identifies the data after a firm accepts the final offer, a step the article says takes 15 to 30 days.

The general buyers publish their own screens. We read each program page on November 10, 2026.

None of those figures is a quote for a CPA firm. The only published professional services comparison, per Polyshares' track record page, is the $110,000 marketing agency license, which involves none of the confidentiality constraints above. Set expectations against that rather than against the $3M ceiling. Every buyer in this market is read side by side in which AI company pays the most for business data.

What should a firm do when it receives one of these offers?

A firm receiving one of these offers should establish three things before it discusses money, and the AICPA guidance's own checklist starts in the same place: "Establish the firm's disclosure and licensing rights."

First, what the buyer is actually asking for. Ask in writing whether the proposed scope reaches client information in any form, including redacted or aggregated. The answer decides whether the conversation continues on your terms or stops.

Second, what you hold that is unambiguously yours, which is usually the methodologies, templates, internal workflow records and training material. Describe only that.

Third, the terms. Some licenses are perpetual or irrevocable, and a trained model cannot unlearn what it has been trained on, so the grant you sign is the last word on that material. Polyshares describes each of its four published licenses as a "Perpetual exclusive license to the operational record". Price the term and the exclusivity separately from the data, and read the sublicensing clause, because a non-exclusive license that permits onward distribution can reach further than an exclusive one. Our page on exclusive versus non-exclusive licenses covers how to split them, and the rights and privacy rules that sit under all of this are in is it legal to sell your company data to AI companies.

Then take it to counsel, which the AICPA guidance advises for the Sec. 7216 analysis specifically. BLACKSIG does the commercial half of this with firms: establishing what is licensable, saying plainly when the answer is very little, and taking the ones that qualify to the buyer whose terms fit.

Frequently asked questions

Related resources

• Can I sell my company's data to AI companies?
• Can a law firm license its data to AI companies?
• What happens to your data after you license it
• How BLACKSIG helps companies license their data