Glossary
Transparency obligation: the AI Act duty to inform
A transparency obligation is a legal duty under the EU AI Act, chiefly Article 50, requiring providers and deployers of certain AI systems to inform people that they are interacting with AI, that content was AI-generated or manipulated, or that an emotion recognition or biometric system is operating.
Illustration · PNG file
- Signature + IHDR
- iTXt XML:com.adobe.xmprecord
- tEXt parameters
- tEXt prompt / workflow
- tEXt invokeai_metadata
- eXIf
- caBX C2PA manifest
- IDAT pixels
- IEND
AI source type declared
The file labels itself as AI-made.
IPTC DigitalSourceType such as trainedAlgorithmicMedia is read from XMP and shown with its exact value.
Check an image →Provider and deployer duties
Providers owe the interaction notice of Article 50(1) and the machine-readable marking of Article 50(2). Deployers owe the emotion and biometric notice of 50(3), the deepfake label of 50(4) and the disclosure of AI-generated public-interest text unless it passed human review with editorial responsibility. Article 50(5) requires every disclosure to be clear and distinguishable, no later than the first interaction or exposure, and accessible.
The duties are cumulative: a company that both builds and deploys a generator owes both sets.
How isGenAI evidences them
The Article 50 scanner captures disclosure text, its placement and timing on public products. The C2PA checker and AI image detector inspect generated files for machine-readable marks. Scorecards record each observation as verified, partial, not observed or unverifiable, with evidence hashes, so a disclosure claim can be checked rather than asserted.
Beyond the AI Act
Platform rules add their own transparency duties: TikTok and YouTube require creators to label realistic AI content, and Meta applies labels from metadata. These are contractual, not statutory, but they shape what a downloaded file and its post reveal.
Documentation is part of the job. A provider that can show the marking technique it chose, the outputs it tested and the export paths that preserve the mark is in a far better position than one that can only point to a policy statement.
How to use this term when checking a file
- Upload the original file to the AI image detector or the tool named on the page.
- Find the finding that uses this term under “How we checked” and read the source field it names.
- Weigh it by kind: a signed record, a declared field, a saved prompt and a matching copy are different strengths of evidence.
- Remember that absence of a record is not evidence of a human author.
Frequently asked questions
Which article creates transparency obligations?
Article 50 of the EU AI Act, applying from 2 August 2026, with definitions in Article 3.
Is marking output a provider or deployer duty?
Machine-readable marking (50(2)) is the provider’s; visible deepfake disclosure (50(4)) is the deployer’s.
How do I prove I met the obligation?
Keep the disclosure copy, placement and timing records, and inspect real outputs for the mark; isGenAI’s scanner and checkers produce evidence you can cite.
Do platform labels satisfy Article 50?
They are separate platform rules. The statutory duties fall on the provider and deployer regardless of platform behaviour.
Sources
Definitions follow the published standards linked above; the isGenAI behaviour described is the deployed checker as of October 2026.