EU AI Act Article 50 transparency checklist: role, marking, disclosure and evidence
A copyable Article 50 preparation checklist for mapping AI roles and surfaces, separating machine-readable marking from visible disclosure, assigning evidence owners and preserving questions for responsible legal/product review.
No email gate. Copy the inventory and checklist below as preparation for responsible legal/product review. It does not determine whether Article 50 applies to a specific system, replace legal advice or constitute evidence of compliance.
Feature inventory
Map the role and surface before choosing the disclosure control.
Article 50 contains different transparency duties. Keep direct interaction, generated-content marking, visible disclosure and other scope questions as separate fields instead of reducing everything to one “AI label” checkbox.
AI feature / surface | Provider, deployer or mixed role | Direct interaction? | Generated or manipulated content? | Machine-readable marking evidence | Visible disclosure decision | Exception / scope question | Evidence owner | Review owner | Last reviewed
Copyable checklist
Evidence first, then the publication decision.
- [ ] Every relevant AI feature or publication surface is inventoried separately rather than treating the whole company as one AI use case - [ ] Provider, deployer or mixed role is recorded for each feature and left for qualified review where the role is unclear - [ ] Direct human-AI interaction is identified separately from generated or manipulated content publication - [ ] Generated-content type is recorded: text, audio, image, video or another supported output - [ ] Machine-readable marking or provenance evidence actually present in the output is recorded without inferring authorship from absence - [ ] Visible disclosure requirement and publication wording are reviewed separately from machine-readable marking evidence - [ ] Deepfake, public-interest text, emotion-recognition or biometric-categorisation questions are flagged for the applicable Article 50 review path rather than collapsed into a generic AI-label rule - [ ] Any relied-on exception, obviousness assessment or transition question is documented with its source and responsible reviewer - [ ] Disclosure timing, placement and accessibility are reviewed for the actual user exposure surface - [ ] Evidence owner and human publication/release owner are named before the feature or content is shipped - [ ] Last-reviewed date and current Commission guidance/source are retained with the checklist - [ ] Checklist completion is treated as preparation for review, not legal advice, automatic applicability determination or compliance certification
A provenance result is evidence, not the legal conclusion.
The European Commission states that Article 50 transparency obligations apply from 2 August 2026 and distinguishes different duties for providers and deployers. Use the current official guidance for scope and exceptions rather than assuming every AI feature has the same disclosure rule.
Machine-readable marking evidence and visible disclosure are separate controls. No C2PA manifest does not prove human authorship, and a present manifest does not by itself determine every Article 50 obligation.
AI Transparency Pack can support the evidence side for a bounded batch of supported media using pinned C2PA inspection plus human disclosure context. It does not decide legal applicability, generate content, detect AI authorship or certify AI Act compliance.