If a prompt names a customer, the GDPR is probably in the request too. It decides who is responsible, what contract you need and where the data may go.
How it works
Regulation (EU) 2016/679 applies “from 25 May 2018” (Article 99). A few definitions do most of the work for API users:
- Personal data (Article 4(1)): “any information relating to an identified or identifiable natural person”, including names, identification numbers, location data and online identifiers. See PII.
- Processing (Article 4(2)): any operation on personal data, such as collection, storage or use. Sending a prompt to a model is processing.
- Controller and processor (Article 4(7) and (8)): the controller “determines the purposes and means”; the processor “processes personal data on behalf of the controller”. You are usually the controller and the LLM provider your processor.
Scope reaches beyond the EU. Article 3(1) covers processing by an EU establishment “regardless of whether the processing takes place in the Union or not”, and Article 3(2) covers non-EU organisations that offer goods or services to people in the EU.
The articles that matter most for an LLM API:
| Article | What it requires |
|---|---|
| 5(1)(c) | Data minimisation: only what is necessary |
| 28 | A binding DPA and rules for each subprocessor |
| 32 | Appropriate security |
| 44 to 49 | A legal basis for transfers outside the EEA, such as adequacy or standard contractual clauses |
| 83(5) | Fines up to 20,000,000 EUR or 4% of total worldwide annual turnover, whichever is higher, for some infringements |
The UK GDPR is the UK’s version, overseen by the ICO. The European Commission lists the UK among adequate countries (renewed on 19 December 2025), so data can flow from the EU to the UK without further safeguards.
Why it matters
- Prompts are processing. A support bot that sends a customer’s message to a model processes personal data, even if nothing is stored.
- Location is a transfer question. If the GPUs are outside the EEA, Chapter V applies; see data residency.
- Retention is only one part. Zero data retention helps with storage limitation, not with contracts or transfers.
The GDPR checklist for LLM APIs turns these articles into 15 questions for any provider.
On BroadNet Inference
BroadNet’s privacy statement does not mention the GDPR by name or a DPA. It says inference content is not retained, inference data is not used to train, fine-tune or evaluate models, and BroadNet’s own and hosted models run on BroadNet-owned or leased hardware in the United States, with partner capacity as backup. A subprocessor list is available on request.
So a request from the EU is a transfer to the US. Ask BroadNet for its DPA, subprocessor list and transfer basis through the contact page before you send EU personal data, and lock keys to your servers. This page is not legal advice.