Stop Putting Client Information Into Your Paid ChatGPT or Claude Account

Accountants paste in client tax documents. Consultants upload strategy decks. HR teams summarize employee complaints. Lawyers draft from case files. Many of them do it in a ChatGPT Plus, ChatGPT Pro, Claude Pro, or Claude Max account, on the assumption that a paid account is a professional one.

A paid personal account is a consumer account with a credit card attached. The subscription buys more usage and better models. It does not buy a different contract, a data processing agreement, or any promise that your client's information stays confidential. Anything you put into one of these accounts goes in on the same terms as a free user.

This article walks through what OpenAI's and Anthropic's own legal documents say, why it matters for anyone who handles client information, and what to use instead. A later section covers the specific ethics rules and court decisions for lawyers.

Both companies say this plainly in their own documents.

OpenAI. The ChatGPT Terms of Use apply to "ChatGPT, DALL·E, and OpenAI's other services for individuals." Paid subscriptions sit inside those same Terms, under a section called "Paid accounts" that covers only billing, renewals, cancellation, and price changes. Nothing in it changes how your content is used. The same document says a separate set of Business Terms governs ChatGPT Enterprise, the API, and OpenAI's other services for businesses (OpenAI Terms of Use).

Anthropic. Anthropic's Consumer Terms cover Claude.ai and Claude Pro, and Anthropic's own offer terms state that the Consumer Terms apply to Pro and Max plans, while the Commercial Terms apply to Team and Enterprise plans (Anthropic offer terms, Consumer Terms). The Commercial Terms say the reverse just as clearly: consumer offerings such as Claude.ai are governed by the Consumer Terms instead (Commercial Terms).

So the line is not free versus paid. It is personal versus business. A $200-a-month Claude Max or ChatGPT Pro subscription is on the same contract as a free account.

OpenAI's Terms of Use say it may use your content "to provide, maintain, develop, and improve our Services," comply with law, enforce its policies, and keep its services safe. If you don't want your content used for training, the Terms tell you to opt out in your settings (OpenAI Terms of Use).

Anthropic's consumer plans work the same way. Since late 2025, Claude Free, Pro, and Max chats are used for training unless the user turns it off, and opted-in data can be kept for up to five years (Anthropic).

Turning training off helps, but it is not the same as a contract:

It is a setting, not a promise. The vendor can change its defaults and its policies. An individual employee can forget to switch it off, or switch it back on.

It has exceptions. On both services, clicking thumbs up or thumbs down can send the whole conversation for training even with training off. On Claude's consumer plans, chats flagged by safety systems can still be used to train Anthropic's safety tools.

It does not make the vendor answerable to you. There is no confidentiality clause covering your input, no processing instructions, and no deletion obligation you can enforce. Your client's information is simply "Content" under a consumer agreement.

A data processing addendum (DPA) changes the vendor's role. Instead of a company that may use your data for its own purposes, it becomes a vendor bound to process your data only to serve you, under your instructions.

Anthropic's Commercial Terms, which cover Claude Team and Enterprise, state that Anthropic may not train models on Customer Content, that Customer Content is the customer's Confidential Information, and that data is processed under Anthropic's DPA, which is incorporated by reference. They also require Anthropic to destroy confidential information on request, subject to legal and backup exceptions (Commercial Terms).

OpenAI's Business Terms (now titled the OpenAI Services Agreement) cover ChatGPT Business, ChatGPT Enterprise, and the API, and OpenAI offers a DPA for all three (OpenAI Business Terms, OpenAI enterprise privacy). Business data is not used for training by default (OpenAI business data).

The consumer terms were written for individuals, and a firm that runs client work through them inherits terms built for an individual. A few examples from OpenAI's Terms of Use (OpenAI Terms of Use):

• Liability is capped at the greater of what you paid in the prior 12 months or $100.
• Businesses indemnify OpenAI. If you use the consumer service as a business or organization, you agree to indemnify OpenAI against third-party claims arising from your use and your content.
• Disputes go to individual arbitration, with a class action waiver, unless you opt out within 30 days.
• Work email accounts can be absorbed. An account created with a firm email address may be moved into the organization's business account, where an administrator can then see its content.

Anthropic's Consumer Terms have a similar provision: an account opened with an employer's email address may be linked to that organization's enterprise account, and its administrator may be able to monitor and control it (Consumer Terms). A personal account opened with a work email address is less separate from the business than it looks.

Most professionals who handle client information have promised, by contract, rule, or law, to keep it confidential. A consumer AI account sits outside all of those promises.

Client contracts and NDAs. Most restrict sharing confidential information with third parties unless they are bound by equivalent protections. A consumer AI provider with the right to use your content to improve its services has no such obligation to your client.

Healthcare. HIPAA requires a business associate agreement before a vendor handles protected health information. Personal plans do not offer one, and OpenAI's business terms prohibit health information unless a separate healthcare addendum and business associate agreement are signed first (OpenAI Business Terms).

Tax and accounting. Tax return preparers generally may not use or disclose tax return information without the taxpayer's consent under Internal Revenue Code section 7216, and CPAs owe confidentiality under the AICPA Code of Professional Conduct. Uploading a client's return to a consumer AI account is hard to square with either.

HR and employment. Employee medical records, investigation files, and personnel data are often protected by state privacy laws and company policy.

In every one of these settings, your client will hold you responsible for where their information went.

Moving off personal accounts is the first step. Self-serve business plans still keep chats until someone deletes them, and they run on standard terms the vendor can change. When a court ordered OpenAI to preserve ChatGPT chats in the New York Times copyright lawsuit, business-plan (then Team) users were covered by the order; Enterprise and zero-retention customers were exempt.

That is why large organizations negotiate enterprise contracts and zero data retention. Even zero retention still sends client data to the vendor's servers. The only approach that does not depend on a vendor keeping its word is making sure highly sensitive information never leaves your business.

Rule 1.6(c). Lawyers must make reasonable efforts to prevent unauthorized disclosure of client information. Choosing a tool whose contract lets the vendor use that information to improve its products is hard to defend as reasonable when a business plan with contractual protections costs little more.

ABA Formal Opinion 512. The ABA advises informed client consent before putting confidential information into self-learning AI tools, and warns that boilerplate in an engagement letter is not enough (ABA Formal Op. 512). Consumer plans are self-learning by default.

United States v. Heppner (S.D.N.Y. 2026). The court held a defendant's chats with consumer Claude were not privileged, and pointed to the consumer terms as a reason the communications were not confidential (Proskauer). The terms you accept are evidence.

Morgan v. V2X, Inc. (D. Colo. 2026). The court barred confidential discovery material from any AI tool unless the provider is contractually prohibited from training on it and from disclosing it beyond what the service requires, with a contractual right to delete and written documentation of those protections. A training toggle in a consumer account's settings is not a contractual prohibition. The judge was explicit about who this leaves out:

• Stop entering client information into any personal ChatGPT or Claude account, free or paid, starting today.
• Move everyone to at least a business plan. Claude Team and ChatGPT Business start at 2 seats, so even a small practice qualifies. For the most sensitive work, look at enterprise terms with short retention or zero data retention.
• Confirm the DPA is in place. For OpenAI, complete its DPA form; for Anthropic, the DPA is incorporated into the Commercial Terms. Keep a copy on file.
• Turn on the shortest retention period your plan allows, where available.
• Close or stop using personal accounts created with work email addresses, and route everyone through the organization's admin-managed workspace.
• Adopt a written AI policy that names the approved tools and bans personal accounts for client work.
• Remove client names and identifying details before anything goes into any AI tool, business plan or not. What never leaves your firm cannot be trained on, preserved, or produced.
• Tell clients how you use AI, in engagement letters or contracts, and get consent where your professional rules require it.

A paid personal subscription buys you more AI, not more protection. The contract that governs your client's information is the same one a free user clicks through. Business plans with a DPA are the minimum for client work, and even they are only part of the answer: courts can still order chats preserved and produced, which is why keeping identifying details out of AI tools altogether is the strongest safeguard.