A data breach through an AI chatbot: a plan for the first 72 hours
Short answer
If an employee put personal data into a public AI chatbot, data protection authorities such as the Dutch AP treat that as a data breach. You have 72 hours from discovery to assess whether to notify. Establish what was shared, contain the damage (delete the chat, ask the provider to delete), assess the risk to the people concerned, notify your supervisory authority if there is a risk, and record everything in your breach register.
Hours 0 to 4: establish and contain
- Respond without blame. An employee who reports it gives you the best chance to act in time. Thank them.
- Establish what was shared. Which tool, which account (personal or business), which data, about how many people, and when? Take screenshots of the chat before it is deleted.
- Contain the damage. Delete the chat and any uploaded files. Check whether the account keeps history or uses data for training, and switch that off. With a business contract, ask the provider in writing to delete the data.
- Inform the DPO and, if it is also a security incident, your security team.
Hours 4 to 48: assess the risk
Not every breach must be notified. The question is whether it is likely to result in a risk to the rights and freedoms of the people concerned. Weigh, among other things: When using ChatGPT counts as a data breach is covered in Is ChatGPT GDPR-compliant?
| Factor | Higher risk if… |
|---|---|
| Type of data | It concerns health, criminal records, financial data or a national ID number |
| Scale | It concerns many people, for example a whole customer file |
| Identifiability | Names, addresses or file numbers were included |
| Control | You cannot have the data deleted, for example with a personal account |
| Vulnerability | It concerns children, patients or other vulnerable groups |
Hours 48 to 72: notify and record
- Notify your supervisory authority within 72 hours of discovery if there is a risk (Article 33 GDPR). If you do not know everything yet, notify what you know and add to it later.
- Inform the people concerned if the risk is high (Article 34 GDPR), in plain language: what happened, what the consequences may be and what they can do.
- Record it in your breach register, also if you do not notify, with your reasoning.
- NIS2: if your organisation is in scope and it is a significant incident, an early warning within 24 hours also applies. See NIS2 and AI.
Afterwards: prevent it happening again
A data breach through AI is almost never malicious. It is an employee looking for help with their work. The Dutch city of Eindhoven found in 2025 that it was happening at scale. See Shadow AI. Structural solutions:
- an approved AI environment where staff can do the same tasks, including with sensitive data;
- safeguards that stop ID numbers, IBANs and file numbers before they reach a model;
- blocks on public AI services, once the alternative is in place;
- an AI policy and training. See Writing an AI policy.
With werqly that environment runs on a node of your own in your server room. What is entered there does not leave your organisation: no outside AI provider gains unauthorised access, so there is no data breach of this kind.
Frequently asked questions
Questions on this topic
Is putting personal data into ChatGPT always a data breach?
According to the Dutch Data Protection Authority it is when an employee does so on their own initiative and against company agreements: the provider then gains unauthorised access. Whether you must notify depends on the risk to the people concerned.
What is the deadline for notification?
Within 72 hours of discovering the breach, if it is likely to result in a risk to the people concerned. If you do not have all the information yet, notify what you know and add to it later.
Can I ask OpenAI to delete the data?
With a business contract you can request deletion, and retention is often agreed. With an employee’s personal account the employee can delete the chat, but the organisation has no contractual grip.
Should the employee be punished?
Usually that backfires: if reporting is punished, the next incidents stay hidden. Focus on removing the cause: an approved alternative, clear rules and training.
Sources
- Dutch Data Protection Authority — Caution: use of AI chatbot may lead to data breaches (2024)
- GDPR (Regulation (EU) 2016/679), Articles 33 and 34
- Municipality of Eindhoven — Data breach involving public AI (press release, Dutch)
This article is general information, not legal advice. Rules and guidance change; check the sources listed or ask your lawyer or DPO if in doubt.