AI Act and transparency of AI generated content

Transparency is more than one ‘generated by AI’ label. An obligation depends on the company's role, use of the system and type of content. Implementation therefore starts with an inventory and verification of the current legal requirements.

Build a use-case inventory

For each process, we record the owner, provider, audience, input data, output type and publishing location. We distinguish a system interacting with a person from an internal tool, and synthetic content from material merely assisted by AI.

Determine the role and obligation

A company may act as a provider or deployer. Article 50 of the AI Act covers, among other matters, informing people when they interact with AI and marking certain artificially generated or manipulated content. Exceptions and application dates require case-specific assessment.

Documentation: EUR-Lex — AI Act, Article 50.

Place information in the right context

A notice should appear where the recipient makes a decision or encounters the system. It should be understandable and accessible. We do not hide it only inside lengthy terms when the context calls for direct information.

Preserve provenance evidence

We retain the tool version, material sources, extent of human editing, publishing decision and labelling method. For product imagery, we also verify fidelity to the real item; an AI label does not correct a misleading presentation.

Review changes

A new model, publishing channel or feature can change the assessment. The process owner reviews the inventory, notices and documentation. Website information does not replace case-specific legal advice for a particular deployment.