EU AI Act
Under Regulation (EU) 2024/1689, the AI features of Entropy Data are classified as limited risk. They fall into no Annex III high-risk category: they govern enterprise data, not employment, creditworthiness, essential services, biometrics, or law enforcement.
Entropy Data is the provider of the AI system in every setup: we place it on the market under our name. You are its deployer — you use it in your own name, for your own purposes. We do not train or fine-tune models, so no general-purpose AI model obligations apply to us.
Roles by setup
The classification is the same in all three setups. What changes is who operates the model and where prompts go.
| Setup | Who deploys the model | Where prompts go |
|---|---|---|
| Cloud + managed model | Entropy Data | The model we operate, under our contract and data-processing terms |
| Cloud + bring your own model | You | Your model endpoint, under your contract. We never see the model provider relationship |
| Self-hosted + bring your own model | You | Your model endpoint, from your own infrastructure. Prompts never touch our systems |
In all three you remain the deployer of the AI system, so deployer duties — human oversight of the features you enable, and telling your users when they interact with AI — are yours. Bringing your own model adds the model side: its terms, its region, and its retention become yours to govern.
Self-hosting adds one further point. We have no visibility into your instance, so audit logging, model choice, and which AI features are switched on are entirely under your control.
Human oversight
Most AI features are advisory. The output is presented to a human — the data consumer, data product owner, data steward, or contract author — who reviews and confirms before any change takes effect.
Purpose-based Access Control is the exception. It decides at query time, without a human confirming each decision. It stays low-risk because it can only restrict access and never widen it, and because it denies rather than allows when the model is unavailable.
All AI features are optional and inactive until a model is configured. See AI Configuration.