How to review AI transparency tasks under Article 50
Short answer
Start with your operator role, then review direct interaction, synthetic output, deepfakes, public-interest text and biometric or emotion exposure. Treat unknown facts as unresolved and inspect each finding together with its reason, next step and official source.
- Establish your roleA provider develops or has an AI system developed and places it on the market or puts it into service under its name. A deployer uses an AI system under its authority. You can perform both roles; integrating another provider does not automatically resolve your own role.
- Answer using actual product behaviorDescribe what users encounter and what outputs are produced. Unknown is a useful answer when you have not checked a feature. A declared notice or marking setting remains self-reported until tested.
- Review exceptions explicitlyFor direct interaction, assess whether AI use is obvious in context. For public-interest text, the editorial exception requires both human review or editorial control and a person or organisation holding editorial responsibility. Special editing, artistic and law-enforcement situations need separate review.
- Export the findings and collect evidenceThe report retains your answers, reasons, next steps and official source. Select relevant cases in the evidence checklist separately; the link does not transfer answers. Capture the actual implementation and revisit it after material changes.
A narrow screen does not classify the entire system
Employment, credit, education and essential-service decisions are prompts for further review. The checker does not determine high-risk or prohibited status, perform a data-protection assessment, or cover the complete AI Act. An implemented status means you reported the control, not that Utilumo inspected it.