Article 50 of the EU AI Act separates two roles. A system provider must make synthetic output machine-detectable where required; a professional deployer must also inform people clearly in specified situations. For a campaign, metadata and visible disclosure are therefore two different design problems.
What began applying on 2 August 2026?
Article 50 transparency duties now cover direct AI interaction, synthetic content, emotion recognition or biometric categorisation, and deepfakes. Providers must inform people when they directly interact with AI in relevant cases and mark synthetic audio, image, video or text in a machine-readable, detectable form.
Deployers carry a different responsibility. When people encounter a qualifying deepfake, emotion or biometric system, or certain AI text on matters of public interest, disclosure must be clear, distinguishable and made no later than first exposure. These duties are not confined to systems classified as high-risk.
There is one limited grace period: systems placed on the market before 2 August 2026 have until 2 December 2026 only for Article 50(2) machine-readable marking and detection. Content generated before 2 August is not retroactively covered by these duties.
Two layers of transparency: technical provenance and human notice
A machine-readable mark may sit in metadata, watermarking or another detection mechanism so platforms and verification tools can identify synthetic material. Providers are expected to make the solution effective, interoperable, robust and reliable as far as technically feasible; standard editing that does not substantially change input or meaning is excluded.
An ordinary viewer is not expected to inspect metadata. Where a deployer must disclose a deepfake, notice needs to be visible or audible. A file emerging from a compliant tool does not replace a clear explanation in a caption, end card, artwork note or campaign page.
An AI label should not be a patch applied after the campaign; it belongs in the original decision about trust, tone and audience experience.
Does every creative image need a large label?
No. The law does not automatically treat every retouch, lighting correction, dust removal or assistive AI use as a deepfake. Where evidently artistic, creative, fictional or satirical work contains a qualifying deepfake, its generated or manipulated nature still needs disclosure, but the notice may be designed in a way that does not obstruct display or enjoyment.
The difficult boundary appears when a synthetic image functions like evidence: a model who does not exist, a false appearance by a real person, an unmade product presented as a sales photograph, or a voice attributed to someone specific. The question is no longer only aesthetic. What belief about reality is the viewer being asked to form?
AI text and the value of editorial responsibility
AI-generated or manipulated text published to inform the public on matters of public interest requires disclosure unless it has undergone human review or editorial control and a natural or legal person holds responsibility for publication. A spelling correction is not the same as substantive editorial judgement.
For a journal or brand, the exception is not a licence for automated publishing. It increases the value of a named editor, fact-checking, retained sources and a responsible owner. If nobody can explain where a claim came from or why it was published, a stronger model does not solve the accountability problem.
A practical workflow for brands and studios
Create a simple AI-asset register: tool and version, date, principal inputs, asset rights, nature of modification, approving person and publication location. Preserve originals and metadata, and decide where a human-facing notice is required. Contracts with photographers, models, voice talent and studios should address model training, identity simulation and permitted uses explicitly.
Ask four questions for each output: will viewers treat it as reality; does it imitate someone's identity; is public-interest material being published without meaningful human review; and will the output be supplied or used in the European Union? The legal answer depends on context and jurisdiction; this article is editorial analysis, not legal advice.
Verdict: transparency can become a mark of brand quality
A weak brand treats disclosure as an admission of failure. A strong one treats it as part of an artefact's provenance. When an AI image has been directed carefully, a short, well-designed explanation of how it was made can clarify the boundary between lived event, concept study and advertising without diminishing the work.
European law sets a floor; brand trust can set a higher ceiling. Even where a specific label may not be mandatory, concealing origin when it changes customer interpretation can create reputational cost. The useful question is not how small a disclosure can become, but what a viewer needs to know to judge fairly.
Frequently asked questions
Must every AI-edited image be labelled?
No. Standard editing or assistive use that does not substantially alter the input or its meaning can be excluded. The content, degree of change, use and likelihood of deception matter.
Is metadata alone enough when a deepfake must be disclosed?
No. Machine-readable marking is a provider duty; in covered situations the deployer must also give people clear, distinguishable notice at first exposure.
Does the EU rule automatically govern every Iranian brand?
Not every domestic use. If a system, service or output is placed on or used in the EU market, territorial scope and each party's role require legal assessment.
Historical, technical and regulatory facts were checked against official and primary sources. Vendor performance claims are attributed to their publishers and kept separate from editorial analysis. This is not legal, investment or environmental advice.
- European Commission — Guidelines on Article 50 transparency obligations
- EU AI Act Service Desk — Article 50 reference page (non-consolidated text)
- European Commission — Code of Practice on Transparency of AI-generated Content
- European Commission — AI Act enforcement and transparency requirements from 2 August 2026
- European Commission — Questions and answers on Article 50