Legal edition

In a criminal case, the examining magistrate at the Rotterdam district court ruled that confidential, privileged information can lose its protected status once someone enters it into ChatGPT. The case involved a person under seizure who used a prompt to draft a reply, apparently meant for a privileged professional. The generated text also included that person's name. The examining magistrate reasoned that an external AI system like ChatGPT can store both the input and the generated text, and use it to train the model. According to the magistrate, this means the communication is no longer confidential. For lawyers, notaries and other privileged professionals, this is the first Dutch signal that using a public chatbot is not just a disciplinary risk. It can also affect privilege in an actual case. Privilege protects communication between a client and a privileged professional. The reasoning is that you break that confidentiality yourself if you hand the text to a third party that stores it and may reuse it. So it is not about what the AI answers, but about where the text ends up. The answer counts too: the name of the privileged professional appeared in the generated text, and so became part of the stored data. This is a ruling in one case, not a decision by the Supreme Court. So do not assume this line is now fixed. But the risk is one-sided: you gain little from using a public chatbot, and seized data cannot be undone. The line lies in the storage and training terms of the tool. A setup that contractually guarantees input is not stored or reused stands on different ground than a free consumer account, but you must be able to prove that difference.

Source →