What does a business plan actually give you?
Three main things, all of them valuable: a contractual commitment not to train models on your data, administrative control (who uses it, under which settings), and a legal framework — a DPA that lets you meet GDPR requirements vis-a-vis the vendor. Compared with a consumer account, where conversations may be used to improve models depending on settings, that is a genuine upgrade.
So what's missing? Contractual vs. architectural protection
A contractual promise says: 'we hold your data, and we commit to treating it a certain way'. An architectural guarantee says: 'your data never reached us'. The difference shows up exactly where contracts don't: a security incident at the vendor, a bug that crossed content between accounts, an employee pasting a client file into a personal account, or a future policy change.
| Question | Business plan (contractual) | Anonymization (architectural) |
|---|---|---|
| Does data leave the organization? | Yes — to the vendor's servers | Identifiers never leave at all |
| What protects it there? | Contract terms and vendor security | Nothing to protect — the data isn't there |
| A breach or leak at the vendor? | Exposure is possible | Nothing to expose |
| Depends on policy changes? | Yes — must be monitored | No |
Does Enterprise mean you can paste anything?
No. Even with a DPA, feeding personal data to an external tool is processing and transfer that require a legal basis, data minimization and security under GDPR. And the list of what never to paste into ChatGPT — health records, legal case files, payroll data — remains relevant on a business plan: the contract governs your relationship with the vendor, not your duties to clients and employees.
What does the right setup look like with a business plan?
- The business plan as the foundation — corporate accounts for everyone, and a ban on running work material through personal accounts.
- Document classification — what may be pasted directly, and what requires anonymization first.
- Anonymization as a standing step — documents containing personal data get cleaned: consistent token replacement (PERSON_001) and metadata removal, before upload.
- Policy monitoring — AI vendors update their terms; assign an owner who reviews the data-usage policy periodically.
The two layers then work together: the contract governs what must be sent, and anonymization ensures that what is truly sensitive is never sent at all.