Why a premium personal subscription to Claude or ChatGPT is not a data governance decision, and why the Team, Business and Enterprise tiers are where client work belongs.
I get some version of the same qauestion every few weeks, usually from a smart executive who has been using AI heavily and has started to wonder what happens to everything they have typed into it. It usually sounds like this: “I pay for the top plan, so my data is private, right?” The honest answer is no, or at least not automatically, and the gap between what people assume and what the terms say is wide enough to drive a regulatory finding through.
The assumption makes intuitive sense. In most of the software we buy, paying means you are the customer and not the product, and the premium tier comes with better protections. With the leading AI assistants, that logic breaks down. Price buys you more capacity, better models and more features. It does not change which set of terms governs your data. What changes that is whether you are on a consumer plan or a commercial one.
What the Consumer Terms Say
Anthropic and OpenAI both draw a hard line between their individual plans and their business plans, and the expensive individual plans sit squarely on the consumer side of that line.
Claude (Free, Pro and Max) -
Anthropic will train new models on data from Free, Pro and Max accounts when the model improvement setting is turned on, and that includes Claude Code sessions run from those accounts. If you leave the setting on, retention extends to five years; if you turn it off, you stay on the standard 30-day retention period. Even with training disabled, conversations flagged for safety review can still be analyzed, and anything you submit through the thumbs up or thumbs down buttons can be used. Max is a consumer plan, full stop, regardless of what it costs each month.
ChatGPT (Free, Plus and Pro) -
OpenAI is more direct about the default. On Plus, Pro and Free personal workspaces, data sharing is enabled by default, and you have to go into Settings, then Data Controls, and switch off “Improve the model for everyone” to stop it. Opting out only applies to new conversations going forward, so whatever you discussed before you flipped the toggle is already in the pool. OpenAI also notes that if you give feedback on a response, the entire conversation tied to that feedback may be used for training, even if you have opted out.
The practical takeaway is simple. On both platforms, the premium personal plans leave the privacy decision to the individual user, one account at a time, through a toggle that most people never look at. That is a user preference, not a control.
|
|
Claude Pro/Max |
ChatGPT Plus/Pro |
Claude Team/Enterprise |
ChatGPT Business/Enterprise |
|
Training on your data |
Only if the user enables it |
On by default, user can opt out |
Not by default |
Not by default |
|
Who controls the setting |
Each individual user |
Each individual user |
The organization |
The organization |
|
Contract posture |
Consumer terms |
Consumer terms |
Commercial terms, customer as data controller |
Business terms and enterprise privacy commitments |
|
Central admin and visibility |
None |
None |
Yes |
Yes |
Why the Business Tiers Change the Equation
The commercial plans are not just the consumer product with a different logo. They sit under a different contract, and that contract is what a security program can stand on.
Training is off by default - Anthropic does not use data from its commercial products, Claude for Work and the API, to train its models by default. OpenAI says the same for ChatGPT Business, Enterprise, Edu and its API platform, and it extends that commitment to data pulled in through connected apps and agent sessions. The burden shifts from a hundred individual users remembering a toggle to a default the organization can rely on.
The organization owns the data relationship - Under Anthropic’s commercial terms, the customer is the controller of the data its users submit and Anthropic acts as the processor, processing that data on the customer’s instructions. That is the same controller and processor framing you already use with your other cloud vendors, and it is the framing your clients, auditors and cyber insurers expect to see when they ask how AI fits into your third-party risk program.
You get governance, not just privacy - The business tiers give you a central workspace, administrative control over who has access, and the ability to offboard someone and keep the work product. OpenAI points out that ChatGPT Business data is encrypted in transit and at rest, and the enterprise tiers on both platforms add identity integration and retention controls. None of that exists when your staff are each running a personal account on a corporate card, or worse, on their own.
It closes the shadow AI gap - This is the part that keeps me up at night more than the training question. If your organization has not provided a sanctioned business tier, your people are almost certainly using personal accounts anyway, and they are pasting in client names, contract language, financials and the occasional screenshot of something that should never leave the building. A business tier gives you a sanctioned path, and a sanctioned path is the only realistic way to shrink the unsanctioned one.
The cost argument is weaker than people think - A single premium personal plan can run higher per month than a standard business seat. When an executive tells me the business tier is too expensive, I usually find they are already paying more for less protection.
What I Tell Clients to Do
The fix is not complicated, and none of it requires you to slow down adoption.
- Decide where client and regulated data is allowed to go - If the answer is “an AI assistant,” then the answer has to be a business or enterprise tier under commercial terms, not a personal account.
- Move the people who handle sensitive work onto Team, Business or Enterprise seats - Start with leadership, finance, HR and anyone touching client deliverables, then expand.
- For anyone who stays on a personal plan, turn training off today - On Claude, that is the model improvement setting under privacy. On ChatGPT, that is Settings, Data Controls, “Improve the model for everyone.” Remember that on ChatGPT the change only covers new conversations.
- Write it down - An acceptable use policy that names the approved tools and tiers, says what data may and may not go into them, and explains why, does more to change behavior than any technical control you can buy.
- Add your AI vendors to your third-party risk register - Review their terms the way you would any processor that touches client data, and revisit them at least annually, because these terms have changed several times in the past two years and will change again.
The Bottom Line
Paying for a premium AI subscription buys you a better tool. It does not buy you a data protection agreement, it does not put your organization in control of the setting that matters, and it does not give you any visibility into what your people are sharing. Those things come with the commercial tiers, and for any organization that handles client or regulated information, that is where the work belongs.
If you are not sure which plans your team is on, or what has already been shared through personal accounts, that is a good place to start the conversation. It is a quick fix, and getting it right now is a lot cheaper than explaining it to a client later.
Looking to learn more about how AI can help your business? Reach out today to get started.
