A risk-tier classification for every AI-touching feature in the product (prohibited / high-risk / limited-risk / minimal), with the reasoning documented rather than just the label, and Article 50 disclosure copy ready for the consent screen or Terms of Service: one always-visible sentence, a collapsed detail block using a native details/summary element, and the human-review invariant stated so it satisfies both the AI Act carve-out and DSGVO Art. 22 at once. Two checklists are included: a plain-language stakeholder version and a shorter sign-off list for the owner.
What ports as-is
- The four-tier classification walk (prohibited / high-risk / limited-risk / minimal)
- The two settlement questions: autonomous decision with legal or significant effect about a person, and the human-review invariant as a product invariant enforced in code
- The Article 50 two-tier disclosure pattern using a native details/summary element
- Terms of Service AI-use clause copy
- A timeline table marking which EU AI Act deadlines apply to this product and which do not
- The plain-language stakeholder checklist
- The owner sign-off checklist
What we build for you
- The list of AI-touching features and the model or provider behind each
- Disclosure wording in the product's own language and register
- The consent or Terms of Service surface the copy goes into
- Counsel review before any external compliance claim is made
