AI & Automation

“I Have a Business AI Account.” What Happens to Your Clients’ Tax Data?

/
10
min read
Lee Reams, CEO of CountingWorks PRO
Lee Reams
Founder & CEO | CountingWorks PRO

A tax professional recently told someone on our team that uploading client documents to an AI assistant wasn't a concern.

“I have a business account.”

I understand why that feels like the answer. They chose a business product, understood that it offered privacy protections, and felt comfortable using it.

But a client's tax return contains more than the information needed to answer a question. It can include Social Security numbers, dependent information, income, investments, and bank account details.

Before that document gets uploaded, the firm should know what the service will receive, why it needs that information, and how it will be handled.

A business account can provide important protections. It does not replace the firm's responsibility to make those decisions.

The shortcut happens during busy season

It's a Tuesday night in March. A staff accountant is three returns behind.

A client's Schedule K-1 has an entry she doesn't recognize. The partner is busy, so she opens an AI assistant, uploads the document, and asks for an explanation.

The response arrives in seconds. She has enough information to continue her research and moves on.

But what else went with that question?

The client's name? Taxpayer identification number? Address? Information about the partnership that wasn't relevant to the issue?

The assistant may have needed the box, code, amount, and surrounding facts. It may not have needed the complete document.

Uploading the whole file was simply easier than finding and removing every unnecessary detail.

The employee was trying to get work done. The firm still needs a process for protecting the information involved.

If that process depends entirely on someone remembering to redact a document during busy season, it deserves a closer look.

Read: AI Is Creating a New Bottleneck: Human Judgment

Your clients didn't hire you to share their tax returns with an AI company

In January 2026, a federal judge upheld an order requiring OpenAI to produce 20 million de-identified ChatGPT conversation logs as evidence in copyright litigation involving The New York Times and other publishers.

OpenAI challenged the order, arguing that it raised significant privacy concerns for people whose conversations had nothing to do with the lawsuit. The company ultimately complied.

There is an important distinction here. The records came from consumer ChatGPT conversations created between December 2022 and November 2024. OpenAI has stated that the production did not involve ChatGPT Business, Enterprise, Edu, or API customers.

The records were also subject to de-identification and legal confidentiality protections. They were not simply released to the public.

Those details matter. This wasn't an example of a business AI provider indiscriminately publishing confidential customer information.

But it demonstrated something worth understanding.

Information shared with an AI service may become subject to legal discovery, even when the provider objects to turning it over.

Think about what that means for a tax professional uploading a client's financial records.

A provider's promise not to train AI models on customer information is valuable. But it doesn't answer every question about storage, retention, access, or disclosure.

And removing a name doesn't necessarily make financial information impossible to identify.

A tax return can contain combinations of income, investments, businesses, property transactions, and family circumstances that reveal a great deal about someone, even when direct personal identifiers have been removed.

The question isn't simply whether an AI company trains on your data.

It's what information you're sending, who can access it, how long it's retained, whether it can be disclosed through legal process, and whether you're authorized to share it in the first place.

Read the January 5, 2026 federal court ruling

Read OpenAI's response to the litigation

Start with the rules that already apply to your firm

Tax professionals already work under requirements governing client information. Those requirements still matter when a new tool enters the office.

Internal Revenue Code Section 7216 restricts tax return preparers' disclosure and use of tax return information.

Unauthorized disclosures or uses can carry criminal penalties under Section 7216 and civil penalties under Section 6713.

The FTC Safeguards Rule requires covered tax preparation firms to maintain a written information security program and oversee service providers handling customer information.

For your firm, that means understanding where client information goes, assessing the risks, selecting appropriate providers, controlling access, and training your staff.

An AI subscription does not complete that work.

If employees are uploading client documents, AI needs to be addressed in your written information security plan, or WISP, and in the procedures your team actually follows.

That includes determining which tools employees can use, what information they can submit, and what safeguards must be in place before that information leaves the firm.

Yes, AI is software. That doesn't settle the disclosure question.

Tax firms have used software to process sensitive information for decades.

Using AI does not automatically require separate taxpayer consent for every task.

The regulations under Section 7216 permit certain uses and disclosures without separate consent, including specified tax preparation and auxiliary services. Other activities require consent.

Treasury Regulation § 301.7216-2 establishes conditions for permitted disclosures and distinguishes certain processing activities from substantive determinations or advice affecting a taxpayer's reported tax liability.

That is why neither blanket answer works:

“You always need consent to use AI.”

“It's software, so consent never matters.”

A tool transferring information onto a return and an external service analyzing how tax law applies to a client's circumstances may raise different questions.

The recipient, contractual arrangement, purpose of the processing, and where the information is processed can also matter.

A firm needs to evaluate the actual service and how it is being used.

Where consent is required, it must meet the applicable requirements. General permission to use technology should not automatically be treated as sufficient.

What a business AI account does—and what it doesn't do

Commercial AI products can offer meaningful contractual and security protections.

OpenAI and Anthropic state that they do not use commercial customer inputs and outputs to train their models by default, subject to their applicable terms and exceptions.

That's valuable.

But not using information to train a model is different from not storing the information.

It is also different from determining whether the firm was authorized to disclose that information in the first place.

For example, Anthropic's commercial retention documentation describes different retention arrangements for API requests and saved product conversations, along with exceptions.

There is another distinction that professionals sometimes miss.

Paying for an AI subscription does not necessarily mean you're using a commercial business product.

Anthropic identifies Claude Pro and Max as consumer offerings, separate from Claude for Work and its commercial API.

The firm needs to understand the exact product and workspace being used, not simply whether an employee pays for it.

And even with an approved commercial account, the firm still needs to understand how that provider handles information.

A business account is one part of a data protection strategy. It is not the entire strategy.

Related: The IRS Just Drew a Line in the Sand on AI. Here's What Every Tax Professional Needs to Know.

Four pitfalls your team should understand

1. An approved account doesn't authorize every use.

A firm might approve an AI product for researching tax questions or drafting client communications.

That doesn't automatically mean employees should upload complete tax returns, financial statements, and supporting documents into every feature of that product.

Approval should identify what staff may do with the tool and what client information they may submit.

2. No training doesn't mean no storage.

Model training, data retention, access controls, and legal disclosure are separate issues.

A provider may restrict model training while retaining certain information to deliver the service, comply with legal requirements, or address security concerns.

Review each issue rather than relying on a single privacy promise.

3. Removing an SSN doesn't remove every obligation.

Tax return information is broader than direct personal identifiers.

Financial facts, business details, and other information supplied in connection with preparing a return may remain protected even when names and taxpayer identification numbers have been removed.

Redaction can reduce exposure. It does not automatically remove Section 7216 obligations.

4. A policy only helps if people follow it.

A partner using an approved business workspace doesn't protect documents a seasonal employee uploads through a personal AI account.

That is why employee training, approved workflows, and accountability matter.

These aren't reasons to stop using AI.

They're reasons to take the same disciplined approach to AI that firms should already apply to other systems handling sensitive client information.

How we approach this in MAX

When we built AI into CountingWorks MAX, we started with a question:

How do we give tax professionals the benefits of AI without asking them to abandon their responsibility to protect client information?

Our approach addresses three important concerns: what information reaches external AI models, how the firm maintains visibility into client activity, and how client consent is handled.

1. Minimize sensitive information before it leaves the platform

MAX is designed to process tax documents within our platform to identify and remove sensitive personal identifiers before sending the resulting information to external AI models for analysis.

This reduces reliance on employees manually finding and redacting every document.

That's especially important during tax season, when staff members are handling large volumes of information and working under deadlines.

The goal is to provide the AI model with the information needed to perform the task without unnecessarily transmitting identifying details.

External AI models may still perform the analysis. The safeguard comes from reducing the sensitive information included in what they receive.

Automated detection has limits. Removing identifiers doesn't eliminate every privacy risk or remove the firm's legal obligations.

That's why data minimization needs to work alongside provider arrangements, access controls, firm policies, and consent where required.

2. Make client activity visible and auditable

Protecting client information isn't only about what happens when a document is uploaded.

It's also about maintaining accountability for the work being performed.

MAX includes client-level activity feeds that maintain a history of activity performed within the platform.

This gives tax professionals a place to review recorded client activity and better understand the work completed on a client's behalf.

That matters when questions arise.

What work was performed? What actions were taken? What happened in the client's workflow?

Instead of relying entirely on someone remembering what they did weeks or months earlier, the firm has an activity history it can review.

An activity feed isn't a substitute for comprehensive security monitoring, and it doesn't necessarily record every operation performed by an external AI provider.

But it provides an important layer of visibility and accountability within the practice.

Good security isn't just about restricting access to information. It's also about knowing what happened.

3. Address AI consent within the client workflow

For client-facing AI capabilities in ClientHub, we've introduced a separate AI consent step rather than relying solely on general terms and conditions.

This helps bring the use of AI into the client experience instead of leaving it as an invisible part of the process.

A firm must still determine whether its particular use of AI requires taxpayer consent under Section 7216 and whether the consent obtained meets the applicable requirements.

A product acknowledgment should not automatically be assumed to satisfy every legal consent obligation.

The goal is to give firms practical ways to address privacy, visibility, and client authorization as part of the work they're already doing.

These safeguards don't eliminate every risk.

But they reflect an important principle: responsible AI use should be built into the workflow, not left entirely to individual employees to figure out.

What to put in place before the next upload

Your firm should be able to follow a short, specific process.

1. Use approved tools and workspaces. Identify the exact AI products, account types, and activities employees are permitted to use.

2. Limit the information submitted. Send only what the task requires. Use an approved process for removing unnecessary identifying information before documents reach external AI services.

3. Review the provider. Understand how it handles model training, data retention, access, processing locations, subcontractors, and legal requests.

4. Address AI in your WISP. Assign responsibility for approvals, employee access, training, service-provider oversight, and incident response.

5. Evaluate authorization and consent. Determine whether a permitted exception applies or compliant taxpayer consent is required for the actual activity.

6. Maintain appropriate activity records. Understand what your systems record, what they don't record, and how your firm can review client activity when questions arise.

7. Train with real examples. Show staff what to do with a K-1, an IRS notice, or a client email. Make it clear when they should remove information, use a different workflow, or ask for guidance.

Someone should own this process and keep it current.

AI products change. Provider terms change. Your firm's workflows will change, too.

Read: The PHP Security Wake-Up Call: Is Your Tax & Accounting Firm Really Protected?

Be ready to explain it to your client

Would you be comfortable with someone uploading your complete tax return to an AI service without explaining what they were sharing?

Your answer might depend on the service, the purpose, the safeguards, and whether the identifying information was necessary.

Your clients deserve that same consideration.

If a client asks what happened to their tax information, the firm should be able to explain which services received it, what information was shared, why it was needed, and what protections applied.

It should also know what records are available to help account for the activity.

“We have a business account” is only part of that answer.

Your clients trust you with their financial information. That trust doesn't become less important because AI is doing some of the work.

Before you upload another complete tax return to an AI assistant, ask two questions:

What information does this service really need?

And can my firm account for how that information was handled?

Those are questions every tax professional should be able to answer.

See how CountingWorks MAX helps firms minimize sensitive client information, maintain visibility through client activity feeds, and put responsible AI workflows into practice:

See MAX in Action.

FREE ASSESSMENT

Would Google Recommend Your Firm Today?

Google’s AI-powered search is changing how firms get found online. Discover what’s helping—or hurting—your visibility, authority, and growth potential.

SAMPLE FINDING

Your firm is visible. It isn’t memorable.

Google can find your website. But neither Google nor prospective clients can quickly explain what makes your firm different from competitors in your market. In an AI-powered world, that’s becoming a serious growth problem.

Take the Free Assessment

Takes minutes • Expertly curated report

Lee Reams, CEO of CountingWorks PRO
Lee Reams
Founder & CEO | CountingWorks PRO

Lee Reams is the founder and CEO of CountingWorks, an AI-native platform helping independent tax and accounting firms build more profitable, less owner-dependent businesses. His focus is practical: use AI to reduce routine work, strengthen client relationships, and help professionals deliver expertise clients value and pay for. With more than twenty years serving the profession, Lee brings a firm owner’s perspective to how technology should work—starting with the result the firm wants to achieve.

Continue Reading...