Can I enter company data into ChatGPT? Risks and safe alternatives

Whether you may enter company data into ChatGPT depends on two questions: what kind of data it is and what type of account you use. General content that does not relate to individuals is usually unproblematic; customer data, employee data and trade secrets, on the other hand, do not belong in private or free accounts where input may be used for training. There is no blanket approval – but there are clear rules of thumb that provide certainty in everyday work.

In brief

• The risk is not AI use as such, but which data ends up in which account.
• Personal data is subject to the GDPR – entering it into third-party systems requires a legal basis.
• With free and private accounts, input may be used for AI training by default.
• Business and API access exclude use for training and come with contractual commitments.
• The safest option for company documents is an AI based on your own documents, with EU hosting and a data processing agreement (DPA).

What determines whether an input is allowed?

The answer comes from the combination of data type and account type – the same input can be harmless in one case and a breach in another. A general draft text without names in an Enterprise account is something completely different from a customer list in a private free account.

When it comes to the type of data, three levels can roughly be distinguished. Uncritical is general content with no link to individuals and no confidential value, such as help with wording or publicly available information. Sensitive is internal information such as cost calculations, strategies or draft contracts. Strictly protected is personal data – that is, any information that can be attributed to an identifiable person, from a customer’s address to an employee’s sick leave.

When it comes to the type of account, the most important dividing line runs between private accounts on the one hand and business or API access on the other. The differences mainly concern use for training and contractual assurances – more on this below.

What are the risks with personal data?

The GDPR (General Data Protection Regulation, the data protection law applicable throughout the EU) only allows personal data to be processed on a legal basis and only for specified purposes. Anyone who enters customer data into an AI service passes it on to a third party – and this requires, among other things, a data processing agreement (DPA: the written agreement that binds the service provider to instructions and protection obligations).

With private ChatGPT accounts, exactly this basis is missing. There is no data processing agreement between your company and the provider, no purpose limitation and no control over where the data is processed. Entering customer or employee data into such an account is therefore regularly a data protection breach – regardless of whether anything ever “happens”.

The consequences can be serious: the data protection authority can impose fines for breaches, the people affected can demand access to information and compensation, and the reputational damage with customers often weighs more heavily than both. Reliable information on obligations and the rights of data subjects is provided by the Austrian Data Protection Authority at dsb.gv.at.

Important for context: this section gives a general overview and does not replace legal advice in an individual case.

What happens to input in free accounts?

With free and private ChatGPT accounts, input may be used by default to improve the AI models, unless users actively opt out in the settings. Your texts then become part of the material from which future model versions learn.

For trade secrets, this is a risk in its own right, separate from data protection. Cost calculations, formulations, customer lists or strategy papers lose their legal protection as trade secrets if the owner does not take appropriate confidentiality measures – entering them into a system without a confidentiality commitment is hard to defend as an appropriate measure.

There is also a practical problem: the sheer lack of control. Once entered, data can neither be retrieved nor specifically deleted, and nobody in the company knows later which information ended up in which private account and when. Shadow AI – the quiet use of private accounts for company purposes – is therefore the real risk in many companies.

What do Team, Enterprise and API access do differently?

The business versions of ChatGPT and API access (API: the programming interface through which software providers build AI models into their own applications) differ from private accounts in three decisive respects. These differences are the reason why many companies restrict AI use to such access.

Feature	Free/private account	Team/Enterprise/API
Use of input for training	Possible by default	Contractually excluded
Data processing agreement (DPA)	Not provided	Available
Management by the company	None	Central user and permission management
Confidentiality commitments	Hardly any	Contractually agreed

A business account thus solves the most serious contractual problems – but not automatically all questions. The server location and the specific transfer routes to third countries still need to be checked, and the limits in terms of content remain: even the most expensive account does not know your company’s files and does not answer questions about your documents more reliably than a free account.

In short: business accounts make use cleaner contractually, but not automatically more useful for document-related questions. For the use case “answers from your own documents”, a different approach is needed.

Which rules of thumb help employees in everyday work?

Experience shows that prohibitions alone work poorly – a few memorable rules that allow productive work are better. Five rules of thumb have proven their worth:

1. No names, no numbers: Personal data – names, addresses, customer numbers, health data – has no place in third-party AI services.
2. Enter nothing that could not also go on a notice board: If you would hesitate to pin a piece of information to the notice board, do not enter it into a private AI account either.
3. Anonymise rather than do without: The real customer case becomes a neutral example – “a plumbing and heating company” instead of the company name, invented figures instead of real cost calculations.
4. Only use approved access: For company purposes, use only the account or tool provided by the company, never a private one.
5. When in doubt, ask: A named contact person for AI questions lowers the threshold and prevents people quietly going it alone.

In addition, a short written AI policy that records these rules is worthwhile. One page is enough – what matters is that everyone in the team knows it and that there is an approved tool they can switch to with a clear conscience.

What safe alternative is there for company documents?

For questions about your own company documents, the safest route is an AI that works on your own document base rather than on a third-party chat account. Such systems – technically usually based on RAG (retrieval-augmented generation: the AI first searches for the relevant passages in the stored documents and answers only from them) – keep the data in a controlled environment and back up every answer with the source in the original. The various routes, with their pros and cons, are compared in the guide Using ChatGPT with your own data.

On the data protection side, this makes it possible to establish exactly what private accounts lack: EU server location, a data processing agreement (DPA) with the provider, contractual exclusion of any use for training and user management that controls access by role. The AI knowledge base from Goma-it is built on these principles; how the rollout works, from the first conversation to training, is shown in the process description.

The side effect of an approved, secure solution is often underestimated: it dries up shadow AI. Anyone in the company who has a tool that answers questions about contracts, quotes and projects reliably and with evidence has no reason left to copy company data into a private free account.