Databricks AI Under the EU AI Act: The Platform Is Not the System
Databricks has no risk tier of its own. The AI system you train and ship is classified by use under Art 6; shipping it makes you the provider.
Databricks itself does not have an EU AI Act risk tier. What you build on it does.
Databricks is a data and machine-learning platform — its Mosaic AI tooling lets you build, fine-tune, evaluate, deploy and monitor models on your own enterprise data, and its Genie feature lets analysts query structured data in natural language. Crucially, this is infrastructure you build on, not a finished AI system someone hands you. Under Regulation (EU) 2024/1689, the regulated object is the AI system, classified by its intended purpose (Article 6) — not the cloud platform underneath it. Mosaic AI is the workshop; the thing you make in the workshop is what the Act assesses.
That distinction has a sharp consequence: when you train a model on Databricks and ship it under your own name, you are most likely its provider (Article 16), carrying the heaviest obligation set in the Act. This page separates the platform from the system, names the roles, and walks a worked example.
Step One: Classify the System You Build, Not the Tool
Using Databricks to run analytics, train an internal model, or let your team query data through Genie carries no mandatory EU AI Act tier for the platform. The classification question only becomes live when you ask: what does the AI system I built actually do, and for whom?
Apply Article 6 to that system:
| What you build on Databricks | Likely tier |
|---|---|
| Internal demand-forecasting or churn model, no effect on individuals' rights | Minimal |
| A customer-facing chatbot grounded on your data | Limited — Article 50 transparency |
| A model that screens job applicants | High-risk — Annex III point 4(a) |
| A creditworthiness / credit-scoring model | High-risk — Annex III point 5(b) |
| Internal data exploration via Genie | Minimal |
A model is not high-risk because it was trained on a capable platform. It is high-risk only if its use lands in Annex III (or it is a safety component of an Annex I product). Most internal analytics and operational models that do not decide anything about a person's rights, access to services, or employment sit at minimal risk, with no mandatory obligations beyond Article 4 AI literacy (in force since 2 February 2025).
If the system does land in Annex III, run the Article 6(3) filter before concluding: a system performing a narrow procedural task, or improving a completed human activity without replacing the human judgement, is not high-risk — unless it profiles natural persons, which is always high-risk. Document whichever way it lands.
Step Two: Building and Shipping Makes You the Provider
This is the part founders underestimate. Under Article 3(3), a provider is whoever develops an AI system (or has it developed) and places it on the market or puts it into service under its own name or trademark. Train a model on Databricks, wrap it in your product, and sell or deploy it under your brand — you are the provider of that system, with the full Article 16 stack if the system is high-risk:
- Risk management system (Article 9)
- Data and data governance (Article 10)
- Technical documentation (Article 11 / Annex IV)
- Record-keeping and logging (Article 12)
- Transparency and instructions to deployers (Article 13)
- Human oversight design (Article 14)
- Accuracy, robustness, cybersecurity (Article 15)
- Conformity assessment (Article 43), EU Declaration of Conformity (Article 47 / Annex V), and EU-database registration (Article 49)
There is also a value-chain flip under Article 25: even if you started by using someone else's system, putting your name on a high-risk system, or substantially modifying it, or repurposing it into a high-risk use, makes you the provider of that system. Fine-tuning a base model on Databricks and shipping it as your own product is exactly this scenario.
A clarification that saves panic: the GPAI duties stay with the model vendor. If your model is built on a third-party foundation model (e.g. an open-weights model you fine-tune), the Chapter V obligations for that general-purpose AI model — technical documentation, downstream information, training-data summary (Article 53) — remain with the original model provider. Fine-tuning can, in some cases, make you a GPAI provider for the modified model, but for most application builders the relevant flip is into a high-risk system provider under Articles 16/25, not into a GPAI model provider.
What Each Role Owes
| Role | Trigger | Core duties |
|---|---|---|
| Deployer (Article 26) | You use a system in your own operations | Use per instructions, human oversight, logging, inform workers where relevant |
| Provider (Article 16) | You build + ship a system under your name | Full high-risk stack: Art 9–15, 43, 47, 49 (if Annex III) |
| GPAI model provider (Chapter V) | You place a general-purpose AI model on the market | Art 53 docs/summary; Art 55 if systemic-risk |
For most companies building on Databricks, the realistic outcome is: deployer for internal tools, and provider for anything you ship to customers. The GPAI layer typically belongs to whoever made the base model you trained on.
The Deadline
If the system you build and ship is stand-alone Annex III high-risk, the obligations apply from 2 December 2027 under the Digital Omnibus — adopted by the European Parliament (16 June 2026) and the Council (29 June 2026), entering into force on Official Journal publication expected before 2 August 2026. (High-risk systems embedded as safety components of Annex I products run to 2 August 2028.) Article 4 literacy and the Article 5 prohibitions are already in force regardless.
Worked Example: Lumen Analytics GmbH
Lumen Analytics GmbH, a German data-products company with ~95 employees and ~€18M turnover, uses Databricks across three workstreams:
- Genie for internal BI — its analysts query the data warehouse in natural language. The AI system here only surfaces the company's own structured data to its own staff: minimal risk, no mandatory obligation beyond Article 4 literacy.
- An internal churn-prediction model trained with Mosaic AI to flag at-risk accounts for the success team — affects internal prioritisation, not individuals' rights or access to services: minimal risk.
- A recruitment-shortlisting model Lumen trains on Databricks and ships inside its own HR SaaS product under the Lumen brand — this is the decisive one. The use is Annex III point 4(a) (selection of candidates), so the system is high-risk; and because Lumen places it on the market under its own name, Lumen is the provider (Article 16), owing the full Art 9–15 stack plus conformity assessment (Art 43), Declaration of Conformity (Art 47) and registration (Art 49). The base model's GPAI duties stay with that model's vendor; Lumen's duties are about the system it shipped. Deadline: 2 December 2027.
Same platform, three very different verdicts — set entirely by what each system does and whose name is on it.
How Confir Helps
Register each AI system you build on Databricks separately — not "Databricks" as one line — and Confir's rule-based engine classifies each by its intended purpose and derives your role (deployer, or provider via the Article 25 flip). For an internal model it returns a short minimal-risk checklist; for a shipped Annex III system it generates the provider obligation set and the Article 11 / Annex IV documentation scaffold. The logic is deterministic: same intake, same finding, with the rule that fired shown in plain language.
Related guides
- Article 6 high-risk classification
- Article 25 value-chain responsibilities
- provider obligations under Article 16
- provider vs deployer explained
- GPAI obligations under the Act
- is my AI high-risk?
- AI system inventory management
Manage your EU AI Act compliance in one place
Confir automates risk classification, technical documentation, and audit trails for any company. No consultants. No 6-month projects. 14-day free trial.
Start free trial →