"Let's deploy an AI chatbot" sounds simple until you realize the chatbot processes your customers' personal data. Then the GDPR applies in full, and since 2025, obligations from the European AI Act are phasing in on top. Here is the concrete checklist we use when designing AIssistant.
The basic frame: who is who
The chatbot operator (your company) is typically the controller of personal data. The AI solution vendor and the LLM API provider are processors. Which means: you need data processing agreements (DPAs) across the whole chain, including the model provider, and you must know where the processing physically happens.
The checklist: what every AI chatbot must have
- A legal basis for processing. Most often contract performance (handling an order/booking) or legitimate interest. Documented, with a balancing test for legitimate interest.
- Transparency. The user must know they are talking to an AI, not a human. Basic decency already required it; the AI Act makes it a duty. Information about data processing must be available the moment the conversation starts.
- Data residency. Where the inputs are processed and where model inference runs. EU processing makes life dramatically simpler; transfers to third countries require SCCs and a transfer impact assessment.
- A DPA with the LLM provider, including a commitment that your data won't be used for model training. Major providers offer this in enterprise terms today, but you need it in the contract, not just in marketing.
- Data minimization. Send the prompt only what's needed for the answer. The chatbot doesn't need the customer's entire record to answer a question about opening hours.
- A retention policy. Conversations must not live forever. Define a retention period, then delete or anonymize. Watch out for backups, erasure has to propagate there within a reasonable window.
- Right to erasure and access. You must be able to find and delete all conversations of a specific data subject, and export them on request. Which means conversations must be discoverable by identity, not just by session.
- Logging and an audit trail. Who accessed what and when, what the AI answered, what action the system performed. Without it you can't meet the 72-hour breach notification duty.
- Human oversight. A path to a human must exist. Suspicious or borderline requests should be escalated, not handled automatically.
- Security. Encryption at rest and in transit, access control, tenant isolation in multi-tenant products, and prompt injection defenses (we cover those in a separate series).
Data residency and LLM APIs in practice
A question that comes up constantly: "Can we even use US LLM APIs?" Answer: yes, but deliberately. Key questions for the provider: Do they offer EU-region processing? Will they sign a DPA? Do they commit to not storing inputs for training? How long do they keep inference logs? For AIssistant we made all-EU processing a hard requirement. It simplifies compliance for us and for our customers.
Right to erasure: the most underestimated item
Erasure sounds trivial until your conversation history is spread across a database, a vector index, logs, and backups. In practice it means: maintaining a mapping of data subject → every storage location, cascading deletes, deleting vector embeddings too (they are derived from personal data), and having a process for backup erasure as backups rotate.
What regulators actually check
Three themes recur in publicly available decisions of European supervisory authorities: transparency (the user didn't know they were talking to an AI, or processing information was missing), legal basis (the chatbot collected more data than necessary), and security (leaked conversations). Have ready: records of processing activities, a DPIA (you'll need one for systematic processing at scale), the DPA chain, and proof that erasure actually works.
Summary
- A chatbot = personal data processing. All of the GDPR applies; the AI Act adds transparency.
- A contractual DPA chain down to the model provider, ideally with EU processing.
- Minimization, retention, discoverability by data subject, and working erasure.
- An audit trail as the foundation: without it you can handle neither an incident nor an inspection.
This is not legal advice, so consult a lawyer for your specific deployment. But if you meet the checklist above, you are far ahead of most deployments out there today.