Are your financial data allowed in ChatGPT & Co.? 5 questions before the prompt.

Somewhere in your company, someone has just copied the latest business analysis, an Excel spreadsheet with contribution margins, or the salary overview into ChatGPT. Of course – it's so convenient… to quickly summarize, check a formula, or have a trend explained.

But what actually happens to these numbers once they leave the text field? What risks do you take when you, as a CFO or in the finance team, work with an AI tool, and what should you pay attention to before entering anything?

The following 5 questions will help you clarify this before the first prompt.

Question 1: Which tariff are you using?

With personal accounts on ChatGPT, Claude, and Gemini, your input is automatically incorporated into the training of the models, regardless of whether you're using the free version or a paid subscription like ChatGPT Plus or Claude Pro. You must actively opt out; otherwise, the model will learn from your input.

Training sounds harmless – but what exactly is it?

This means that your input becomes part of the training material used to train the next version of the model. The model doesn't save your sales list as a file, but its contents influence its future output, and in the worst-case scenario, a fragment of it might reappear in the response for another user.

The business and enterprise plans are different: By default, your data is not used for training, and you receive a data processing agreement. This agreement obligates the provider, in accordance with the GDPR, to process your data only on your behalf and not for their own purposes, including training. For you, it provides the legal basis for being allowed to enter company data into such a tool at all.

For you, this means: The free ChatGPT version, which lets you quickly summarize a sales list – you'd better leave it alone.

Question 2: Have you turned off the training – and have your colleagues too?

So, if you're using a private version, definitely turn off the training. You'll find the button in the settings (in ChatGPT under "Data Control", "Improve model for everyone").

There are two catches: First, disabling it only works in the future: What you entered previously may already be in a training run. Second, your own, correctly set switch is only half the solution. Because there might still be colleagues with their own accounts who never saw the switch.

Therefore, it's worth considering the topic fundamentally: Which AI setup actually suits your company? Which information and figures are so sensitive that they should never be included in an open tool? And which departments that work with it daily are better off using a business or enterprise access?

Question 3: Are you legally allowed to enter financial data?

It depends.

As soon as the numbers you copy into the AI ​​contain personal data, the GDPR applies. Personal data means that the information can be linked to a specific individual. This includes a payroll list with names, a customer list, and applicant documents. An anonymous total in a monthly report, however, does not.

And then the following applies: If an external provider processes such data on your behalf, the GDPR requires a data processing agreement – ​​precisely the contract we explained in question 1. Without it, you lack a legal basis. The input is then not legally permissible, no matter how low the risk may seem in a particular case.

Then there's the question of where the data ends up. Many services process your input on servers in the USA. You also need to be aware of this when dealing with personal data and be able to justify it. Business plans include a contract that addresses these points, but free private accounts do not. Without both, you're ultimately liable.

Question 4: Would you present these figures to a stranger?

Many people imagine AI as a closed machine that no human can look into. This isn't true for the free versions: some of the conversations are read by humans so that the provider can improve its service and train the models.

And even if you've turned off training, OpenAI still stores your input for up to 30 days to check for abuse, and humans can still see it. According to Google, Gemini reviews conversations that have been checked by trained reviewers, including external service providers, and stores them for up to three years. They don't disappear even if you delete your chat history.

Google says this quite openly: In its own Gemini guidelines, it asks users not to enter anything they wouldn't want to see in front of an auditor.

Therefore, it's worth briefly assessing the sensitivity of your figures before inserting them. Are they "harmless" data such as an aggregated industry comparison, a rounded total, or figures that are already included in the annual report? Or are they data that must remain protected? Such as personal data and trade secrets (margin structure, sales pipeline, the purchase price from an ongoing M&A transaction).

What would be damaging to business if it surfaced outside your company? Anything you can think of in that regard shouldn't be left unprotected in an AI text field.

Question 5: Do I really need the actual figures for this?

In most cases, you can use the AI ​​tool without revealing your real values.

Often, all you need from AI is the thinking: explaining a formula, classifying a key figure, drafting a text for the advisory board. The actual figures hardly matter. Invented or rounded values, anonymized placeholders, or a sample instead of the real file are sufficient. Instead of the actual contribution margins, you simply enter example figures of the same order of magnitude, and the answer is just as useful.

Before you add anything, it's worth asking yourself: Does the actual value really need to be included? Often the answer is no, and the risk is eliminated.

The quick check before the first prompt

Go through these five questions before you enter business figures into an AI tool.

1. Which tariff am I using?

2. Is training switched off, and does this apply to everyone on the team?

3. Am I even legally allowed to enter this data?

4. Would I present these figures to a stranger?

5. Do I really need the actual figures for this?

This is the only way you should use AI in finance.

These questions don't mean that AI has no place in the financial sector. On the contrary: it can really help you – but only under one condition…

And that is: a clean data foundation. An AI only ever works with what you give it. If you feed it an Excel file with incorrect numbers, you'll get a quick, well-formulated, but still wrong answer. So the problem isn't with the AI ​​at all – it lies in the underlying data.

That's exactly where it starts. loyos bi Your financial data is collected in our software, where it is automatically processed and made available in clear reports. This gives you access to consistent, audited figures and financial transparency across your entire business at any time. This is the work that no chatbot can do for you, and the foundation on which AI can meaningfully build.

If you would like to know what our Financial Management Suite looks like for a company of your size, we would be happy to show it to you in a personal consultation.

‍