March 23, 2026 | Heinz W. Süess
A colleague is still sitting in the office late at night.
The schedule is full, the HR team is understaffed and the management wants a decision on a sensitive personnel case by tomorrow morning. Out of sheer pragmatism, the employee copies the entire email history, internal notes and a draft justification for the dismissal into a free AI chat.
„Make it legally watertight - and a little more empathetic,“ he types in. What he overlooks at this moment: The free AI is not just a helpful tool. It is a data vacuum (2 Disturbing Ways How AI Chatbots Are Quietly Collecting Your Data, 2025).
Why free AI is rarely really free
Those who work with free AI chats do not pay with money, but with information. The business model of most providers is based on two pillars: Data and training. Every input - from harmless vacation requests to confidential HR dossiers - is stored, evaluated and can be used to train the models. This makes the service smarter, the provider richer in data and the user remains under the illusion that they are getting something for nothing.
What's more, many free versions serve as a „testing ground“. New functions, new models, new evaluation methods - all of this is first tried out on free users. In concrete terms, this means more tracking, more A/B tests, more analysis of your behavior. And the more you use, the more valuable you become. Not because you pay, but because you reveal your organization, your processes and your way of thinking.
Anyone who believes that a free AI is a nice digital colleague is misjudging the distribution of roles. You are not the customer, you are the data source (Zharovskikh, 2025).
What data AI chats collect
AI chats live from content. What you write or upload in the text field is the most obvious part. This includes:
- Texts: e-mails, protocols, contracts, patient case descriptions, staff appraisals.
- Files: PDFs, presentations, Excel lists with personnel data, medical reports.
- Prompts: Including your questions, instructions and internal formulations.
But it doesn't stop there. Metadata and usage data are generated in the background, for example:
- Time, duration and frequency of use.
- IP address, rough location data, device type, browser used.
- Which functions you use, which answers you click on, what you ask for or correct (Zharovskikh, 2025).
A very detailed usage profile can be created from all of this: How you write, which topics you frequently deal with, whether you tend to work with HR cases, medical topics or management decisions. All in all, a pattern emerges that reveals a lot about your company and its internal processes.
In healthcare and HR in particular, these patterns are highly sensitive. When an AI „learns“ how people in your organization talk about performance deficits, illnesses or conflicts, it is more than just „technical training“. It is a look into the inner workings of your organization (ChatGPT Health Launched - ChatGPT In Healthcare: Answers to the Important Questions, n.d.).
What this means for companies and the healthcare sector
For companies and especially healthcare facilities, confidentiality is not a „nice-to-have“ but a duty. Various levels of protection apply here:
- Business secrets: Strategies, financial figures, internal evaluations, HR
- Data protection according to DSG/DSGVO: Personal data of employees, patients and applicants.
- Professional secrecy: Medical confidentiality, nursing confidentiality, psychological care, social counseling (Admogin, 2025).
Anyone who copies such information into a free AI risks several problems at once:
- Loss of control: You do not know where the data is stored, for how long and for what purpose. „Delete“ in the chat window does not necessarily mean that all copies have disappeared.
- Legal risks: If sensitive data is transferred to third parties (i.e. the AI provider) without a clear legal basis, there is a risk of violating the FADP/DSGVO. This is particularly critical in the case of health data and special categories of personal data.
- Reputational damage: A data leak, a misconfiguration of a service, an unfortunate screenshot - and suddenly internal cases are semi-public. Trust can be irreparably damaged, especially in the healthcare sector.
In short:
Free AI can quickly turn a clever shortcut into an expensive data protection construction site.
The good thing is:
You don't have to do without AI to protect your data. But you do need clear guard rails.
How to protect yourself and your team
- No sensitive data in free tools
Make it a basic rule: no names, no personal dossiers, no medical information, no confidential figures in free AI - if you need an example, anonymize consistently: change names, alienate figures, generalize contexts. - Use opt-out and privacy settings
Many providers now offer options to restrict or disable training with your own data - at least in certain plans. Check the settings, switch off everything that is not absolutely necessary and document this for your organization. - Check European or in-house alternatives
Instead of jumping on every hype service, it is worth taking a look at solutions that are operated in Switzerland or the EU in compliance with data protection regulations or even run on your own in-house servers. Models that are operated locally or in a dedicated environment significantly reduce the risk of uncontrolled data leakage. - Clear guidelines for employees
Write a clear, understandable AI policy:- Which tools may be used - and which may not.
- Which data may be inserted - and which may never be inserted.
- How to work with anonymized examples.
Training courses, short checklists and specific „do's and don'ts“ are more helpful than legal novels.
- Define responsibilities
Define who in the company is responsible for selecting and approving AI tools: IT, data protection, HR, management. This will prevent all employees from introducing solutions on their own initiative that end up being a risk for everyone (Becht & Becht, 2026).
In the end, the question is not whether you use AI - but how. Free tools are tempting, but they come with a price that you only realize at second glance. If you know this price, you can consciously decide what role AI should play in your company: Data sucker or tool - victim or creator.
Sources:
2 Disturbing ways How AI chatbots are quietly collecting your data. (2025, April 21). MYDWARE IT Solutions Inc. https://mydware.com/2-ways-how-chatbots-collect-data/
Zharovskikh, A. (2025, May 8). AI data collection guide. InData Labs. https://indatalabs.com/blog/ai-data-collection
ChatGPT Health launched-ChatGPT in the healthcare sector: Answers to the important questions. (undated). Swiss Radio and Television (SRF). https://www.srf.ch/news/ratgeber/chatgpt-gesundheit-lanciert-chatgpt-im-gesundheitsbereich-antworten-auf-die-wichtigen-fragen
Admogin. (2025, September 26). ChatGPT Data Protection Germany: GDPR-compliant use. GIEL | ATTORNEY AT LAW. https://giel-rechtsanwalt.de/allgemein/chatgpt-datenschutz-deutschland/
Becht, J. & Becht, J. (2026, February 19). ChatGPT is not GDPR-compliant. Here are alternatives. WEVENTURE Performance GmbH. https://weventure.de/ch/blog/dsgvo-konforme-chat-gpt-alternativen
