EU AI Act Compliance

BLU DELTA is classified as a "limited-risk AI system" under the EU Artificial Intelligence Act. This classification applies to systems that assist in information extraction without autonomous decision-making. Accordingly, the following principles apply:

• Transparency: BLU DELTA uses machine learning models to extract data from documents. This should be clearly communicated to end users.

• Function Documentation: The system can extract standard document (Invoice, Delivery Note, etc.) fields (e.g., vendor, amount, date) and also supports customer-specific fields. Depending on the configuration, these fields may be defined manually, rule-based, or with the aid of AI components. Extraction accuracy and limitations should be understood before integration.

• Traceability: All model and API versions are documented. Results are traceable and—where applicable—reproducible, depending on the underlying model architecture. Deterministic models provide consistent outputs, whereas models with stochastic elements (e.g., certain LLMs) may yield variable results.

• No Autonomous Decisions: BLU DELTA provides structured suggestions, not final decisions. It does not perform accounting, validation, or legal checks.

• Responsible Use: Integrators are responsible for ensuring appropriate quality control. Fully automated workflows ("dark processing") must include safeguards like verification steps and logging.

We recommend that partners retain JSON responses for audit purposes and clearly disclose to users when AI-extracted data is used.

Disclaimer – Limitation of Liability 

While BLU DELTA aims to provide accurate data extraction, errors may occur due to the probabilistic nature of AI models. The extracted information must be reviewed and validated before use in critical processes. Blumatix shall not be held liable for any damages arising from the reliance on AI-generated results without appropriate human verification. Responsibility for end-to-end correctness lies with the implementing party.