Skip to content

Field NotesHow Accounting Firms Can Protect Customer Data When Using AI

Workflows

How Accounting Firms Can Protect Customer Data When Using AI

Glyph-field title card on dark carbon: dense workflows texture glowing amber, article title "How Accounting Firms Can Protect" on staggered dark slabs.
Accounting firms protect client data when using artificial intelligence by enforcing three controls: signed vendor agreements barring model training, zero-data-retention pipelines that purge prompt logs, and formal Section 7216 disclosure notices or client consents before processing tax records. Treating an AI vendor like any other technical service provider satisfies Federal Trade Commission security rules and keeps client workpapers strictly confidential.

Every managing partner recognizes the immediate temptation of artificial intelligence during tax season. Staff members face hundreds of disorganized client document packages, missing K-1s, and unstructured payroll summaries. Typing a prompt or uploading a ten-page PDF to summarize a business owner's tax posture feels like an obvious shortcut.

The commercial risk, however, doesn't originate from whether modern language models can read a balance sheet. The real danger comes from what happens to sensitive financial figures once they leave your firm's local boundary. When an employee pastes client numbers into a public chatbot or an unvetted cloud platform, your practice hasn't simply adopted modern software. It has initiated an unauthorized disclosure of protected client records.

Protecting your firm doesn't require locking staff out of automation. What it requires is treating every machine learning vendor as an audited service provider. By aligning your data workflows with federal tax statutes, Federal Trade Commission regulations, and professional ethical standards, you can deploy AI tools that accelerate tax preparation while keeping client records completely protected.

The Legal Floor: IRC Section 7216 and Criminal Exposure

Most commercial business owners evaluate software by reading security seals on a marketing page. In an accounting practice, marketing claims aren't a legal defense. Tax preparers operate under a federal statute that predates generative software by decades: Internal Revenue Code Section 7216.1

Under Section 7216 and its civil counterpart, Section 6713, knowingly or recklessly disclosing or using tax return information without explicit authorization is a federal misdemeanor punishable by fines and up to one year of imprisonment. The statutory definition of tax return information isn't restricted to filed Form 1040s or corporate returns. It covers any piece of information furnished to a preparer in connection with preparing a return, including organizer sheets, bank statements, depreciation tables, and client interview notes.

Treasury Regulation Section 301.7216-2 establishes narrow exceptions where disclosures don't require client consent, but those exceptions carry strict conditions.1 Disclosing information to another domestic preparer for auxiliary services is permitted only so long as that third party doesn't perform substantive determinations, such as analyzing, interpreting, or applying tax law to the client's position.1 Furthermore, disclosing data to software and equipment contractors is permitted only to the extent strictly necessary for technical support, programming, or processing, and only if each individual contractor receiving access receives formal written notice that the criminal penalties of Section 7216 apply to them.1

A free consumer chatbot account or self-serve web application doesn't provide statutory penalty notices to its engineering team. Worse, if your staff inputs client facts into an AI tool and asks it whether an expense is deductible or how an S-corporation distribution should be treated, the system performs a substantive determination under the regulatory definitions.1 Without formal consent, that transfer breaks federal law.

The IRS Office of Professional Responsibility underscored this boundary in its guidance on artificial intelligence in federal tax practice (Issue 2026-19).2 The IRS confirmed that ethical duties of competence, diligence, and confidentiality under Treasury Circular 230 remain fully in force when using AI.2 Willful mishandling of taxpayer data through unapproved AI systems exposes practitioners to direct sanction and loss of practice rights before the IRS.2

The FTC Safeguards Rule and Vendor Governance

Federal tax law governs disclosures of return data, while the Federal Trade Commission Safeguards Rule (16 CFR Part 314) governs overall data infrastructure.3 Many accounting firm leaders mistakenly assume the FTC Safeguards Rule applies only to retail banks or mortgage brokers. Under federal law, however, any accounting firm or professional tax preparation business in the practice of completing returns qualifies as a non-banking financial institution.3

The Safeguards Rule requires covered firms to maintain a Written Information Security Plan (WISP) overseen by a designated Qualified Individual.3 When your firm brings an AI platform into its workflow, Section 314.4(f) mandates three explicit duties regarding service providers:3

  1. Taking reasonable steps to select service providers capable of maintaining appropriate administrative, technical, and physical safeguards.
  2. Requiring those safeguards by written contract.
  3. Periodically assessing providers based on the risk they present and the continued adequacy of their controls.

A staff member opening a personal account on a consumer plan provides no contract between the vendor and the firm. If that vendor suffers a credential breach, your firm can't demonstrate compliance with its statutory oversight obligations.

These duties align with the American Institute of Certified Public Accountants Code of Professional Conduct. Under the Confidential Client Information Rule (ET § 1.700.001) and its interpretations for third-party providers (ET § 1.700.040), a CPA must enter into a contractual confidentiality agreement or obtain specific client consent before sharing information with an outside technology provider.4 The AICPA Statements on Standards for Tax Services (SSTS No. 1 § 1.3 and § 1.4) explicitly mandate that relying on software tools doesn't absolve the practitioner of professional responsibility for data protection and work product accuracy.4

Three Deployment Paths and Their Regulatory Exposure

Accounting practices typically adopt generative AI tools through one of three deployment models, each carrying distinct regulatory exposure under federal tax rules and privacy standards.

Comparison of AI deployment models for accounting firms across regulatory, contractual, and operational dimensions.
Deployment modelModel training rightsPrompt retention windowIRS Section 7216 postureFTC Safeguards Rule status
Consumer chatbot accountsPermitted by default across public modelsIndefinite or tied to personal chat historyNon-compliant; unauthorized disclosure of client dataViolates vendor risk management and access rules
Standard commercial SaaSExcluded under terms of service but subject to vendor changeStandard 30-day abuse monitoring logs retainedUncertain; lacks required statutory contractor noticesPartial; requires manual verification of vendor controls
Governed enterprise infrastructureStrictly barred by signed Data Processing AddendumZero data retention or cryptographically isolated logsCompliant via contractor notices or signed client consentFully integrated into firm Written Information Security Plan

Consumer accounts create immediate regulatory exposure, while standard commercial plans leave retention gaps that only governed enterprise agreements close.

Evaluating AI systems across clear operational categories reveals why standard commercial accounts aren't enough for regulated client workpapers.

When choosing between these architectures, firm leaders should look beyond brand recognition. The critical distinction lies in whether data processing occurs under enforceable contractual commitments or informal platform default settings.

Standard commercial SaaS subscriptions often appear adequate at first glance because their marketing pages advertise enterprise-grade security. Yet standard enterprise terms frequently reserve the right to retain prompts and generated outputs for 30 days in centralized logging pools to perform abuse monitoring.5 If that 30-day monitoring window allows vendor safety engineers or offshore contractors to review logged transcripts, your firm has permitted an unauthorized disclosure outside its custody boundary.

Governed enterprise infrastructure eliminates this ambiguity. By routing firm requests through private application programming interfaces with zero data retention enabled, no human reviewer or automated diagnostic pipeline stores client workpapers after a computation completes.

The Technical Architecture of Data Custody

Protecting customer information doesn't require building machine learning models from scratch. It requires establishing technical boundaries around how commercial models receive and process firm documents.

According to CPA.com vendor due diligence standards, an accounting firm must demand specific written representations before connecting any external tool to client files:5

First, verify model training restrictions in the signed agreement. Marketing claims and website FAQ pages don't create enforceable obligations. The vendor's signed contract or Data Processing Addendum (DPA) must state that customer inputs, attached tax files, embeddings, and prompt outputs won't be used to train, tune, evaluate, or improve any foundational or commercial model.5

Second, verify the retention lifecycle for every data asset. You must determine where prompts, extracted tables, and temporary files reside during processing. For client return preparation, your firm should insist on zero data retention (ZDR) endpoints for model inference. When an AI tool reads a PDF bank statement or general ledger export, the system processes the text in temporary volatile memory and purges the data immediately upon delivering the structured response.

Third, examine the vendor's sub-processor chain and server geography. A primary software vendor might host its web interface on domestic cloud servers while routing natural language queries through an external model provider operating data centers overseas. Under Treasury Regulation Section 301.7216-2, transferring tax return information to any server or contractor located outside the United States requires prior signed consent from the taxpayer, regardless of whether that data remains encrypted.1 If the vendor can't guarantee domestic data processing, the tool can't touch tax return files without specific cross-border consent forms.

Fourth, inspect the actual scope of the vendor's SOC 2 Type II report. Many technology startups market a SOC 2 Type II certification that covers only their corporate email and administrative office network, rather than the machine learning infrastructure, model inference pipeline, or storage clusters processing client files.5 Your Qualified Individual or external cybersecurity advisor should examine the auditor's scope description and verify that third-party model providers are evaluated within the audit boundary.

Client Consents and Engagement Letter Updates

Even with enterprise security controls, certain accounting workflows require direct client communication and formal consent under federal regulations.

Treasury Regulation Section 301.7216-3 and IRS Revenue Procedure 2013-14 dictate exact formatting requirements for obtaining valid taxpayer consent.6 If your firm uses AI tools to assist with Form 1040 preparation in ways that require consent (such as performing substantive legal analysis or utilizing offshore processing nodes), the consent can't be buried as a routine paragraph inside your annual engagement letter.6

Revenue Procedure 2013-14 requires that Section 7216 consent for individual taxpayers be executed on a separate written document, printed in at least 12-point typeface on paper, and signed and dated by the taxpayer prior to disclosure.6 The document must include mandatory statutory language in a prescribed sequence, explicitly informing the client that federal law protects their return information and that signing the consent is voluntary.6 Furthermore, conditioning your tax preparation services on the client's signature voids the consent entirely.6

For non-1040 business engagements and general accounting advisory services, update your standard engagement letters to reflect modern technology use. Under AICPA guidelines, informing clients in writing that your practice uses vetted third-party technology providers to assist in service delivery establishes professional transparency and satisfies ethical notice requirements.4

Protecting client data isn't an obstacle to firm efficiency. When you configure enterprise vendor agreements, restrict access through zero-retention pipelines, and document clear security policies, your practice gains the full operational leverage of AI without compromising client trust or violating professional standards.

Frequently Asked Questions

Can accounting firm staff use personal ChatGPT accounts to summarize client documents?

Personal or free chatbot accounts shouldn't ever be used with client data. Consumer AI terms typically allow providers to retain prompts and use submitted information to train future models. Entering client names, financial figures, or tax documents into a consumer account constitutes an unauthorized disclosure under Internal Revenue Code Section 7216 and violates the FTC Safeguards Rule.

What is zero data retention in an AI contract?

Zero data retention means the AI provider doesn't store prompts, uploaded files, or generated outputs on its persistent storage disks once an API request is completed. While standard commercial plans often keep logs for up to 30 days to monitor platform misuse, zero data retention ensures sensitive accounting records exist only in temporary memory during live processing and vanish immediately afterward.

Does an accounting firm need signed consent from every client to use AI?

Signed consent isn't required for every AI application, but it's mandatory in specific regulatory circumstances. Under Section 7216, if an AI tool performs substantive determinations regarding tax law, or if the vendor processes data on servers located outside the United States, individual Form 1040 filers must execute a separate, signed consent form following Revenue Procedure 2013-14. For domestic tools performing auxiliary processing under a signed vendor agreement, standard engagement letter disclosures generally suffice.

Sources

  1. 26 CFR § 301.7216-2 - Permissible disclosures or uses without consent of the taxpayer. Electronic Code of Federal Regulations (eCFR).
  2. Issue Number: 2026-19. Introductory Guidelines for Responsible AI Use in Federal Tax Practice. Internal Revenue Service Office of Professional Responsibility. 2026-07-08.
  3. FTC Safeguards Rule: What Your Business Needs to Know. Federal Trade Commission. 2022-04-27.
  4. Tax ethical standards in data protection and reliance on tools. The Tax Adviser. AICPA Tax Division. 2026-04-30.
  5. AI solution due diligence guide for accounting firms. CPA.com.
  6. Section 7216 Frequently Asked Questions and Revenue Procedure 2013-14. Internal Revenue Service.

Get back tothe work of humans⁠1

Your work should be about creativity, solving problems, and building relationships.
Leave the machine-like tasks to us.