Is ChatGPT safe for accountants? A practical answer

The short answer

ChatGPT is safe for an accounting firm to use, as long as you're careful about which version and what goes into it. With no client information involved, such as drafting a generic newsletter, explaining a standard in plain English or tidying up a spreadsheet formula, the risk is low. With client information, it's safe only on a business plan the firm controls, under terms you've read, written into your security plan, with a decision made on taxpayer consent, and with a person checking everything it produces.

What isn't safe is the version most staff reach for first: a personal account on their own phone or laptop, with client names, figures or return data pasted straight in.

The rules that already apply

There's no special AI law for accountants yet. You don't need one. The rules you already work under cover it.

IRS guidance on AI. In June 2026 the IRS Office of Professional Responsibility published Issue Number 2026-19, its introductory guidelines for responsible AI use in federal tax practice. It says practitioners must review everything AI produces before it goes to a client or the IRS, check facts, citations and calculations, and handle client data only in secure, enterprise-approved AI. It warns specifically against uploading sensitive data to unsecured or public systems, and says wilful mishandling of taxpayer information through AI can lead to discipline under Circular 230.

Section 7216. Preparers face civil and criminal penalties under IRC sections 6713 and 7216 for using or disclosing tax return information without authorisation, unless the client has consented or a regulatory exception applies. The IRS has not said whether sending return information to an AI tool counts as a disclosure to a third party. Until it does, the cautious route is a separate, signed consent before client return data goes to an outside AI provider. For Form 1040 clients, Rev. Proc. 2013-14 sets out the format. The consent has to be voluntary, so you also need a way to do the work without AI for any client who says no.

The FTC Safeguards Rule. Tax preparers and many accounting firms are covered as financial institutions under this rule, which means a written information security plan, controls on who can access customer information, and oversight of the service providers who handle it. An AI vendor that receives client data is one of those providers. A staff member's personal chatbot account has no contract with the firm at all, so it fails that test before you even get to 7216.

Professional confidentiality. CPAs also owe clients confidentiality under the AICPA Code of Professional Conduct and their state board rules. Pasting a client's details into a tool the firm doesn't control is hard to square with that.

Why the plan matters

OpenAI's own pages, checked on 4 November 2026, draw a clear line between personal and business use.

On the personal plans (Free, Plus and Pro), conversations can be used to train OpenAI's models by default. Users can switch that off under "Improve the model for everyone" in their data controls, but it's a setting each person has to change, and the firm can't see or enforce it.

On ChatGPT Business and Enterprise, OpenAI says it does not use your data to train its models by default. The firm owns the workspace, an admin controls who's in it and which connected apps are allowed, and Enterprise adds controls over how long data is kept.

A business plan doesn't settle the 7216 question or write your security plan for you. It does give you a tool the firm actually controls, under terms you can point to. These terms change often, so check the current version before you rely on them.

A simple traffic light for your team

Green: fine on any approved tool. Drafting a generic client newsletter, summarising a public standard or IRS notice, writing an Excel formula from made-up example data, rewording an internal email with no client details in it.

Amber: only on the firm's approved business plan, under your rules. Anything with client names, financials, entity details or correspondence in it. Each use should be one your policy names, with a 7216 decision made for return information, and a person checking the output.

Red: never. Social Security numbers, full returns or source documents in any personal account, client data in any tool the firm hasn't approved, and any AI output going to a client or the IRS without a person checking it first.

Five steps for a small firm this month

1. Find out what's already in use. Ask staff which AI tools they've tried, for what and on which accounts. Make it safe to answer honestly. Our post on finding where your team's AI data is going covers how.
2. Pick one tool and one plan. Approve a business plan the firm controls, read its data terms, and close the gap on personal accounts.
3. Make the 7216 call with your adviser. Decide whether and when you need client consent, draft the form with counsel, and plan the non-AI route for clients who decline.
4. Write it into your security plan. Add the AI tool to your written information security plan as a service provider, with who can use it and for what.
5. Write one page of rules and train the team. Green, amber and red uses, who checks the output, and what to do if client data goes into the wrong place.

An AI Safety Check does steps one and two with you and maps the rest. A Safe AI Setup puts the controls in place. Managed Safe AI keeps them current every month, because IRS guidance, vendor terms and the tools your staff use will keep moving, and someone needs to keep the firm ahead of them.

FAQs

Can accountants use ChatGPT?

Yes. The IRS's own 2026 guidance assumes practitioners will use AI. It expects secure, enterprise-approved tools for client data and a person reviewing everything AI produces before it goes to a client or the IRS.

Is putting client tax data into ChatGPT a Section 7216 disclosure?

The IRS hasn't said. Until it does, the cautious approach is a separate, signed client consent before return information goes to an outside AI provider, using the Rev. Proc. 2013-14 format for Form 1040 clients. Talk to your own counsel about how this applies to your firm.

Does ChatGPT train on our client data?

On ChatGPT Business and Enterprise, OpenAI says it does not train on your data by default. On personal Free, Plus and Pro plans, training is on by default unless each user switches it off.

Do we need to mention AI in our written information security plan?

If client information goes into an AI tool, that vendor is a service provider handling customer information, so it belongs in the plan the FTC Safeguards Rule requires, along with who can use it and for what.

What should we do if someone already pasted client data into a personal AI account?

Write down what was shared, in which tool and when, delete it from the account where you can, and talk to your adviser about whether it needs to be treated as a security event or reported. Then fix the gap that allowed it.

*This post is general information, not legal or tax advice. Vendor terms and IRS guidance change, so check the current versions and talk to your own legal or compliance adviser about your firm's situation.*