Blog / Transparency
Not legal advice. This article is a preliminary technical guide for AI SaaS teams. Always confirm classification and obligations with qualified counsel. Effective dates reflect the Digital Omnibus updates as of July 2026.
The European Commission published its Article 50 implementation guidelines just before the transparency obligations begin to apply on 2 August 2026.
The deadline did not become a blanket obligation to label every use of AI. The correct first step is to identify which Article 50 case applies, which actor carries the duty, and whether an exception or transition rule is relevant.
| Product behaviour | Primary duty to check | Typical actor |
|---|---|---|
| A person interacts directly with an AI system | Inform the person that they are interacting with AI, unless this is obvious in context | Provider |
| A system generates or manipulates image, audio, video, or text | Make outputs identifiable in a machine-readable way where Art. 50(2) applies | Provider |
| Emotion recognition or biometric categorisation is used | Inform exposed persons, subject to the legal exceptions | Deployer |
| Deepfakes or certain public-interest text are published | Clearly disclose that the content was artificially generated or manipulated | Deployer |
These duties are not interchangeable. A chatbot notice does not automatically satisfy output-marking duties, and a website privacy-policy paragraph does not replace a point-of-interaction notice.
List every surface where a natural person interacts with AI or receives generated content:
Record the provider, deployer, intended purpose, audience, launch date, and where the output is used.
For direct interaction, draft clear copy shown at the point of interaction. For generated or manipulated content, separately assess technical marking and detectability under Article 50(2).
The Commission's materials also provide optional EU icon variants for certain labelling use cases. Using an icon does not remove the need to assess the wording, placement, accessibility, and applicable paragraph.
Keep a dated record of:
This evidence is useful for counsel, procurement, and a future request from a competent authority.
Do not describe 2 December 2026 as a general Article 50 extension.
The Commission describes a limited transition for the marking and detection obligation in Article 50(2) for relevant generative AI systems placed on the market before 2 August 2026. It does not automatically postpone chatbot interaction notices, deepfake disclosure, emotion-recognition notices, or every other Article 50 duty.
Confirm system dates and the exact paragraph with counsel before relying on the transition.
For each AI touchpoint, keep one row with:
| Field | Example |
|---|---|
| Surface | Customer-support chat |
| Actor | Provider / deployer / both — confirm |
| Article 50 case | Direct interaction |
| Disclosure | "You are chatting with an AI assistant…" |
| Placement | Before the first AI response |
| Accessibility | Screen-reader label and readable contrast |
| Evidence | Screenshot + release version |
| Owner | Product / Legal |
| Open question | Is interaction obvious in this context? |
Use the Article 50 checklist and disclosure template, or request an Article 50 Evidence Pack for a product-specific implementation file.
Not legal advice. The guidelines support implementation but do not replace the AI Act or product-specific legal analysis.
Get a free readiness brief for your product — not a commodity risk label.
Get a free readiness brief for your product