Product classification under the EU AI Act follows a structured four-step decision tree. Step one verifies whether the software meets the statutory AI system definition. Step two checks if the functionality falls under prohibited practices in Article 5. Step three evaluates whether the system meets high-risk criteria under Annex 1 or Annex 3. Step four identifies active Article 50 transparency obligations, such as chatbot disclosures or synthetic content marking.

An audit trail architecture for AI Act compliance relies on three core technical artifacts. First, an automated logging pipeline records system operational events and model outputs. Second, a living risk register tracks identified hazards and mitigation measures throughout the system lifecycle. Third, technical documentation packages store model architecture specs and evaluation reports for authority inspection.