AI Transparency Obligation

AI Transparency Obligation explained

Under Article 50(1), direct interaction with an AI system generally requires people to be informed unless the AI interaction is already obvious. Put required notices where the encounter begins. A hidden explanation in a legal notice does not reliably explain an ongoing conversation. Design information to be understandable for people with different accessibility needs.

Paragraph 2 concerns machine-readable marking of synthetic outputs by system providers, with legal limits and exceptions such as certain standard editing. Paragraph 4 instead addresses deployer disclosure for deepfakes and certain text publications on matters of public interest. Not every AI output is a deepfake. Classification depends on content and presentation; a technical provenance signal does not automatically replace a required perceptible notice.

For the relevant text publications, the law provides an exception involving human review or editorial control and editorial responsibility. This is not a general exemption for images or videos. Spell-checking alone is not substantive review. As of 11 September 2026, transparency rules generally apply from 2 August 2026; the transition to 2 December 2026 concerns only certain previously marketed systems and paragraph 2.

Creative Engineering connects clear communication with a traceable production process. We take responsibility for the concept and quality. Document what was generated, what review occurred and which notices the actual use requires. Test labels in the published format, including shortened excerpts. Transparency does not make a false statement true or replace cleared rights and factual quality.