1. Where is the real risk?
One of the biggest risks is company information in personal accounts: employees pasting internal documents or client data into free versions of AI, without the company knowing what happens to them.
A personal account and a business plan are not governed by the same terms. In a personal account, the rules on how your conversations are used depend on the settings each person has chosen, and the company does not control them. On a business plan, the terms are set by a contract with the organization, and an administrator manages the accounts.
A business plan does not remove every risk either: connectors, access permissions and shared spaces can also expose information if they are not configured properly.
That is why the first question is not "which tool is safe?" but "which account is it used in, and with what settings?"
2. What AI providers say
The three major providers publish specific commitments for their business plans; these are the ones they state on their official pages.
| Business plan | Model training with your data | Other published commitments |
|---|---|---|
| Claude for Work (Team and Enterprise, by Anthropic) | Inputs and outputs are not used to train the models by default. Exception: if someone explicitly submits feedback (for example, with the thumbs up or down button) or chooses to allow it. | The client company is the "controller" of the data and Anthropic acts as "processor": it only processes the data according to the client's instructions to provide the service. |
| ChatGPT Business and Enterprise (by OpenAI) | "By default, we do not use your business data for training our models," according to OpenAI. | The company owns its inputs and outputs (where allowed by law). OpenAI can sign a data processing agreement (DPA) for ChatGPT Business and Enterprise, to support GDPR compliance. Encryption at rest (AES-256) and in transit (TLS 1.2 or higher). In ChatGPT Business, administrators control retention, and deleted conversations are removed from its systems within 30 days, with the exceptions set out in the provider's policy, including legal obligations and security reasons. |
| Google Workspace with Gemini (by Google) | Google states that it does not use customer data to train models without prior permission, and that content is not used to train Gemini outside your domain or reviewed by humans without permission. | Prompts are treated as customer data under Google's Cloud Data Processing Addendum. Gemini only accesses the Workspace content the user already has access to. |
Sources, accessed on 9 and 10 October 2026: Anthropic — data use for training · Anthropic — controller and processor · OpenAI — enterprise privacy · Generative AI in Google Workspace Privacy Hub. These terms may change: check them on the provider's page before signing up.
3. Personal accounts: Claude, ChatGPT and Gemini
In personal accounts, whether content is used to train the models depends on a setting that each user needs to check.
| Tool (personal account) | What the provider states |
|---|---|
| Claude (Free, Pro, Max) | Training the model with your conversations depends on a setting ("model training") that each user can turn off in their privacy settings. If it is on, new chats may be kept in de-identified form for up to five years for training. A deleted conversation is removed from the systems within 30 days, with exceptions, for example conversations flagged by safety systems or ratings you have submitted. |
| ChatGPT (Free, Go, Plus, Pro) | Content is used to train the models, with an opt-out option. |
| Gemini (personal Google account) | With the "Keep Activity" setting on, Google uses your activity to improve its services, including training generative AI models, with the help of human reviewers; reviewed data is disconnected from your account and kept for 3 years. With the setting off, your future conversations are not used to train the models, unless you send feedback, and they are kept for 72 hours. Feedback you send may also be kept for up to 3 years, without being linked to your account. |
Does an individual paid subscription, such as Google AI Pro, change anything? This is not confirmed in the official documentation consulted. The safe approach is to check the setting in each account.
Sources, accessed on 9, 10 and 13 October 2026: Claude — plans and pricing · Anthropic — terms for Free, Pro and Max accounts · Anthropic — data retention · ChatGPT — pricing · Gemini Apps Privacy Hub
4. What the law requires
Two European texts apply directly to a company that uses AI with data: the GDPR and the EU AI Act.
GDPR, Article 28. If the tool processes personal data on the company's behalf, the company is the controller and the provider is the processor. Article 28 states that the controller "shall use only processors providing sufficient guarantees" and that this processing "shall be governed by a contract or other legal act". For their business plans, OpenAI offers a data processing agreement (DPA), Anthropic states that it acts as a processor under its commercial terms, and Google refers to its Cloud Data Processing Addendum. It is worth checking that the contract has been signed or accepted before entering personal data.
The contract is not the only requirement. The GDPR also calls for a legal basis for the processing, limiting the data to what is necessary and, if the data leaves the European Economic Area, safeguards for that transfer. And the fact that a provider does not train its models with your data does not, on its own, guarantee that you may share it.
EU AI Act, Article 4. Organizations that deploy AI systems must take measures to support their staff's AI literacy. This obligation has applied since 2 February 2025. Regulation (EU) 2026/1744 (the Digital Omnibus), in force since July 2026, keeps it but no longer mandates a specific level of literacy, according to the European Commission. For high-risk systems, the obligation to train the staff in charge of human oversight remains in place.
Sources: GDPR (Regulation (EU) 2016/679), Articles 5, 6, 28 and 44 onward · Regulation (EU) 2024/1689 on AI, Articles 4 and 113 · European Commission — AI literacy, questions and answers, accessed on 13 October 2026
This article is for information only and does not constitute legal advice.
5. Professional firms: an extra level of requirements
For lawyers and advisors, the profession's ethical rules come on top of the GDPR. In Spain, two texts set the reference:
- Circular 3/2026 of the Consejo General de la Abogacía Española (the representative body of the legal profession in Spain). For professional documents drafted with the help of AI, it requires always checking the result: whoever signs a document takes responsibility for its entire content. The circular itself states that confidentiality and professional secrecy will be covered in another circular.
- ICAM good practice guide (Madrid Bar Association). The ICAM guide (presented in October 2025) calls for always keeping human control, adopting internal protocols (authorized tools, people responsible for supervision) and ensuring confidentiality, including with external providers.
The details, with use cases by profession, are in AI for professional firms.
Sources: Circular 3/2026 of the CGAE · ICAM guide
6. Seven practical rules
These seven rules are the ones I recommend for using AI with company data without surprises. They are not legal obligations in themselves, but a prudent way of meeting them.
- A professional environment approved by the company for any use involving internal information or client data.
- Reviewed settings: training, retention and connectors, account by account.
- Real anonymization of sensitive documents before using them: removing names is not always enough. For the AEPD (the Spanish data protection authority), data is only anonymous if there is no reasonable likelihood of identifying the person.
- A list of authorized tools, known to the whole team.
- Prohibited data in writing: data that never goes into an AI tool.
- Human validation of everything that leaves the company.
- An internal champion who answers questions and centralizes needs.
To start with a concrete task and set up your first project, see A practical AI guide for executives.
In our 1:1 training, we apply a simple rule: we always work in the client's AI account, never in ours, and if a document is sensitive, we use an anonymized version. That way we avoid bringing the client's documents into our own account.
7. Frequently asked questions
Is ChatGPT safe for a company?
It depends on the plan. For ChatGPT Business and Enterprise, OpenAI states that it does not train its models with company data by default and offers a data processing agreement. In individual versions, content is used to train the models unless the user opts out.
Is my data used to train AI?
On the business plans of Claude, ChatGPT and Google Workspace, not by default, according to the providers. In personal accounts, it depends on each user's settings.
Do I need a contract with the provider?
If the tool processes personal data on the company's behalf, yes: Article 28 of the GDPR requires that processing to be governed by a contract or other legal act with the provider.
What about the free version?
For internal information or client data, I do not recommend it: its terms depend on each user's settings and the company does not control them. For tests with fictitious documents or documents you are authorized to share, it can work.
What should we do if an employee has already used a personal account with company data?
Identify what information was shared and whether it includes personal data. If it does, it may be a personal data breach: you need to assess the risk and notify the competent supervisory authority (the AEPD in Spain) without undue delay and, where feasible, within 72 hours of becoming aware of it, unless it is unlikely to result in a risk to the rights and freedoms of individuals (Article 33 GDPR). Then review that account's settings, delete the affected conversations if the tool allows it, and give the team a business plan and written usage rules. Consult your data protection officer or your advisor.
Next practical step: organize how your team uses AI with our AI acceptable use policy template for small businesses.
Want to bring order to how AI is used in your company? Book an intro call: 30 minutes, free, and a tailored proposal within 5 business days.