Identifying data
Patient data in a chatbot: a warning from the Dutch regulator
In August 2024 the Dutch data protection authority said it had received several breach reports about employees entering patients' and customers' data into AI chatbots. In one, a GP practice employee had entered patients' medical data.
On 6 August 2024 the Dutch Data Protection Authority (Autoriteit Persoonsgegevens, AP) announced that it had recently received several breach notifications in which employees had shared patients' or customers' personal data with an AI chatbot. Many people use assistants such as ChatGPT and Copilot at work, it noted, for instance to answer customer questions or summarise long files.
One notification came from a GP practice: against agreements, an employee had entered patients' medical data into an AI chatbot. At a telecoms company, an employee had entered a file that included customers' addresses. Medical data are highly sensitive and receive extra legal protection for good reason, the authority stressed; simply sharing them with a tech company is a serious violation of the patients' privacy.
The AP also explained why this is a breach. Most companies behind chatbots store everything entered, so the data end up on their servers, often without the person typing realising it or knowing what will happen to them; the person the data are about will not know either. Entering personal data into a chatbot against the employer's rules is a data breach, which in many cases must be reported to the authority and to those affected.
The authority urged organisations to agree clearly with staff whether chatbots may be used and, if so, which data may go into them. They could also agree with the provider not to store the input. An official told the Dutch public broadcaster that even a general description of an illness without a name can be unwise to enter.
What can we learn from it?
No attacker and no provider failure were needed: one employee trying to save time was enough. That is an understandable aim, and AI really can save a lot of time, if it is settled in advance what it may be used for and with what data. Without such a rule, each person decides afresh, often in a hurry. Confiella sends only reviewed, pseudonymised text to the cloud; pseudonymised text is still personal data, so the judgement stays with the professional. A good first step is a short written rule: which tool may be used, with what data, and what is too sensitive even without a name.
Sources
- Autoriteit Persoonsgegevens: Let op: gebruik AI-chatbot kan leiden tot datalekken
- NOS: Privacywaakhond: pas op met gebruik AI-chatbots, ging mis bij huisarts
This article is based on the public sources listed. If you find a mistake, write to us: [email protected]