An executive speaking six languages in a video, an enthusiastic customer who never existed, an almost photographic product image: AI is blurring the reference points of communication. For brands, the question is no longer simply how to produce better content or cut costs. It is becoming: what should the public be told about how a message was made? Looking ahead to September 2026, the AI Act’s August 2 deadline turns that question into a regulatory and editorial task. With one crucial distinction: not all AI-assisted content is subject to the same obligations.
One deadline, several forms of transparency
The European Union’s regulation on artificial intelligence, adopted in 2024, provides for the transparency obligations in Article 50 to apply from August 2, 2026. The following analysis draws on that text and on trends already documented, without anticipating implementation details or enforcement practices actually observed in September 2026. The operational implications presented here therefore constitute a forward-looking interpretation of that deadline.
The first pitfall is to reduce the framework to a universal obligation to add an “AI-generated” label. The text distinguishes, in particular, between system providers, which must enable certain synthetic outputs to be marked, and organizations that use these systems under their authority, referred to as “deployers.” A brand may fall into this second category when it uses a tool to create and then distribute content.
For providers, Article 50 requires synthetic audio, image, video or text outputs to be marked in a machine-readable format and detectable as artificially generated or manipulated. The requirement takes technical feasibility into account and includes exceptions, notably for certain standard editing assistance functions. This technical marking is not the same as a visible notice: information that software can process does not necessarily inform the reader.
What brands must make visible
For communications professionals, two situations require particular attention. The first involves deepfakes: AI-generated or manipulated images, audio or videos that resemble existing people, objects, places, entities or events and could mistakenly be perceived as authentic. The deployer must disclose their artificial nature. A spokesperson whose voice has been recreated, or a realistic scene fabricated around an event, may fall into this category.
The regulation provides for an adapted approach when such content forms part of an evidently artistic, creative, satirical, fictional or similar work or programme. Disclosure must then remain appropriate and must not hinder the presentation or enjoyment of the work. This does not constitute a general exemption for advertising. A creative campaign cannot automatically assume that its aesthetic is enough to inform the public.
The second situation concerns AI-generated or manipulated texts published to inform the public on matters of public interest. The text establishes a disclosure obligation, but also an exception when the content has undergone human review or editorial control and a natural or legal person assumes editorial responsibility for its publication. Commercial texts are therefore not all automatically subject to labelling.
This exception deserves more than a cursory sign-off before publication. For a brand commenting on health, the environment or a social issue, a prudent approach means being able to explain who checked the facts, decided on the wording and authorized publication. The regulation does not prescribe a single editorial procedure in this article. But merely declaring responsibility, without a corresponding organizational framework, would be a fragile strategy.
Three layers that must no longer be confused
A sound transparency policy separates three complementary functions. Confusing them leads either to a proliferation of unnecessary badges or to the belief that invisible metadata settles the relationship with the public.
- Public disclosure explains what is artificial, in understandable terms and at the relevant moment.
- Technical marking provides tools and platforms with structured indicators of the content’s origin or modifications.
- Editorial responsibility identifies the organization accountable for the message, its accuracy and its consequences.
Article 50 requires the relevant information to be provided clearly and recognizably, no later than the first interaction or exposure. A notice buried in terms and conditions therefore seems difficult to reconcile with this approach. For a video, an on-screen notice legible from the first exposure is one option; for an image, a directly associated caption may be appropriate. These choices must be tailored to the medium and the applicable requirements.
Provenance does not guarantee truth
The industry has not waited for this deadline to explore traceability. The C2PA standard and systems such as Content Credentials allow signed information about a file’s provenance and certain modifications to be attached to it. Several creative and technology companies have begun adopting them. These tools offer promising infrastructure, not a magic solution or automatic proof of regulatory compliance.
A screenshot, a conversion or a republication process can break the continuity of the available information. Above all, knowing a photograph’s provenance does not prove that its caption is accurate. Conversely, synthetic content can explain a phenomenon faithfully. File authenticity, factual accuracy and fair presentation remain three distinct questions. Communications departments must address all three.
The real work starts within teams
In practical terms, a brand should begin by mapping its uses of AI: translation, retouching, dubbing, illustration generation, writing and avatars. It should then assess each case according to the content, its purpose and its potential to cause confusion. Translating an internal draft does not raise the same questions as distributing a fabricated video testimonial. And disclosing that something was artificially created does not negate misleading advertising or an infringement of someone’s rights.
This mapping must extend to contracts with agencies and service providers. Who documents the tools used? Who retains provenance information? Who checks permissions relating to a person’s face or voice? Who adds the final disclosure, including in social media adaptations? Without a clear allocation of responsibility, transparency risks disappearing somewhere between file delivery, cropping and publication.
Wording matters too. “Created with AI” can describe anything from a simple correction to an entirely invented scene. A more precise phrase—“synthetic voice” or “artificially generated scene”—is more helpful when it accurately reflects reality. Good transparency does not showcase every tool used: it reveals what changes the interpretation of the message.
What next? Differentiation may depend less on the presence of a badge than on the consistency of a publicly stated policy: acknowledged uses, explicit limits and an accountable point of contact. As synthetic content becomes commonplace, brands would benefit from making transparency a verifiable promise rather than a defensive reflex. The lasting advantage would not be producing content without humans, but showing where their judgment remains indispensable.
