Le nouveau problème créatif est de prouver que vous êtes humain

Generative AI originally created an authenticity problem because machines became better at imitating people.

Creators are now encountering the opposite problem. People are being mistaken for machines. Business Insider reported in August 2026 that some social-media creators had seen genuine human-made content labelled or treated as AI-generated by platform systems. For influencers whose commercial value depends heavily on perceived authenticity, the mistake can affect reputation and potentially brand relationships.

The episode points towards a strange next phase in generative media. Producing convincing content is getting easier. Proving where content came from is becoming more valuable.

AI quality has created an attribution problem

The first generation of synthetic images was relatively easy to recognise. Hands looked wrong. Text dissolved into nonsense. Faces had an artificial smoothness. Video movements gave themselves away. Those clues have become much less reliable.

Modern generative systems can produce photography, voices, video and written language good enough that visual inspection frequently provides no definitive answer about origin.

At the same time, human creators use AI-assisted features constantly. A photographer might remove an object with generative fill. A filmmaker may use automated noise reduction. A writer can ask software to shorten a headline. A designer can expand an image beyond its original frame.

The useful distinction is no longer simply human versus AI. Creative production now exists on a continuum.

Labels inevitably simplify that continuum

Platforms still need categories. Users want to know when a realistic-looking person never existed. Regulators want safeguards against deceptive synthetic media. Advertisers want confidence that creators are representing their production methods accurately.

Europe has now formalised part of that expectation. Article 50 transparency requirements under the EU AI Act became applicable on 2 August 2026, including requirements concerning AI-generated or manipulated material and specific disclosure obligations for deepfakes and certain other content.

But any labelling system faces a difficult classification problem. How much AI assistance changes the nature of the work?

An entirely generated advertising model clearly differs from a photographer using AI to remove a rubbish bin from the background. Between those examples lie hundreds of ambiguous workflows.

Authenticity could become a technical feature

The industry therefore needs something more useful than guessing from the finished image.Content provenance offers one direction. Instead of analysing an image after publication and asking whether it looks synthetic, software can preserve information about how the content was created and modified.

A camera could authenticate an original capture. Editing software could record subsequent changes. Generative systems could mark synthetic components. Publishing platforms could display selected parts of that history to audiences. The question changes from “Does this look like AI?” to “Where did this file come from?” That is a considerably stronger foundation for trust.

Creators may start protecting originals

The economics could change creator behaviour too. For years, creators primarily worried about other people stealing their work. They may increasingly need evidence that they made it. Original camera files, editing histories, project files, timestamps and provenance metadata can become commercially useful when authenticity is disputed.

Brands may also ask more precise questions. Instead of prohibiting “AI content”, a contract might specify which forms of generation are unacceptable, which editing assistance is permitted and what disclosure must accompany synthetic elements.

That would reflect how creative work is actually evolving. A blanket definition of AI use will become increasingly difficult to maintain as generative functions disappear into ordinary software.

AI detection alone cannot carry the burden

Detection tools will continue to improve. Generation systems will improve too. That creates an adversarial cycle in which one group develops better synthetic media while another tries to identify statistical traces left behind by it.

False positives become particularly damaging in that environment. Calling synthetic content human can spread deception. Calling human content synthetic can damage the person who created it. The second error has received less attention because the industry has concentrated on detecting AI. Creators experiencing incorrect labels show why both sides matter.

Human-made could become a premium category

An unexpected market may emerge from this confusion. Consumers already pay more for handmade furniture, analogue photography, live music and other products partly because human production carries cultural value. Creative media may move in a similar direction.

A publication could certify human photography. A fashion campaign could deliberately advertise that no synthetic models were used. An illustrator might provide provenance alongside commissioned work. AI would still dominate many high-volume production tasks.

Human authorship could become more explicit precisely because it is no longer safe to assume. The irony is difficult to miss. Generative AI spent several years trying to become indistinguishable from human creativity. It is succeeding well enough that humans may now need technology to prove that they are human.