A spectacular image, a familiar voice, a fake political speech, or a viral video will no longer circulate as easily without explanation. Starting August 2, 2026, a new phase of the European AI Act places the transparency of AI content at the forefront. Deepfakes must be labeled, just like certain AI-generated texts on public interest topics when published without human review or editorial oversight. For media, platforms, brands, and creators, the message is clear: the era of invisible synthetic content is coming to an end.
This shift comes at a time when tools capable of producing credible images, voices, and videos are becoming widely accessible. It therefore concerns not only Brussels or the legal departments of tech companies but directly impacts how French and European citizens will judge what they see on their screens, especially during a crisis, an election campaign, a disaster, or a viral controversy.
On August 2, a New Frontier Between Creation and Deception
The European Commission published guidelines in July to assist providers and professional users of AI systems in applying the transparency obligations outlined in Article 50 of the AI Act. These guidelines cover situations where a person interacts with a machine, emotion recognition systems, or biometric categorization, as well as artificial content that may be mistaken for authentic productions.
The case of deepfakes is the most visible. An image, video, or sound generated or manipulated by AI, falsely presenting a person, place, or event as real, must be clearly labeled. Public interest texts produced by AI are also targeted when disseminated without human validation or editorial accountability. The goal is not to ban synthetic creation but to prevent it from masquerading as reality.
Why Media is at the Heart of the Battle
Trust has become as crucial an infrastructure as servers or studios. A false photograph can traverse multiple networks in minutes, be cropped, republished, and then taken out of context. When corrections arrive, the initial impression has often already taken hold. Newsrooms must therefore verify the origin of a file but also explain to the public when artificial intelligence played a material role in its creation.
The Associated Press has notably strengthened its internal rules on July 23. The agency allows AI for certain assistive tasks, such as preliminary research, document summarization, transcription, translation, title suggestions, or coding assistance. However, it maintains a red line: editorial judgment, verification, and accountability remain human. It also continues to prohibit the use of generative AI to create, modify, or enhance news photographs.
This position shows that the debate is no longer simply about âusing or rejecting AI.â The real question becomes: for what task, with what control, what traceability, and what accountability? A tool can accelerate transcription without deciding the truth of an accusation. It can suggest a title without replacing the journalist who verifies the facts and assumes publication.
Content Credentials Become a Trust Weapon
To accompany visible mentions, the industry is working on technical proofs of provenance. The C2PA standard allows for associating information about a file’s origin and the modifications it has undergone. âContent Credentialsâ can thus indicate that an image comes from a compatible device, has been retouched, or was generated by an artificial intelligence tool.
In July, the IPTC, an international reference body for media standards, published a practical guide for newsrooms considering adopting this technology. The timing is not coincidental. Media groups are seeking a way to prove the authenticity of their images just as the public learns to doubt everything. Camera manufacturers, agencies, and tech companies are already testing these mechanisms.
However, technical provenance is not a magic solution. A screenshot, compression, or passing through a platform can strip certain metadata. An authentic file can also be presented with a misleading caption. Labels must therefore complement the work of verification, source research, and journalistic context, not replace them.
Platforms, Brands, and Creators Face a New Risk
European obligations will have global commercial consequences. An American platform, an Asian studio, or an advertising agency based in the Middle East can be affected as long as its service or content reaches the European market. For businesses, the risk is not limited to regulatory sanctions. An advertisement using the synthetic face of a celebrity without labeling can trigger an immediate image crisis.
Creators will also need to distinguish between assumed aesthetics and misleading simulation. An artistic video clearly presented as fictional does not pose the same risk as a fake testimonial, a false press conference, or a voice imitation broadcast as a real recording. The context, visibility of the labeling, and the public’s ability to understand what they are watching will be decisive.
What Will Change Specifically in France
In France, media, businesses, and administrations will need to map the AI uses already present in their production chains. Many organizations use generation, translation, or synthesis tools without a clear public policy. The new phase of the AI Act requires documenting these practices and defining who verifies, who validates, and who is accountable in case of error.
- For newsrooms: maintain real editorial control and explain the material uses of AI.
- For platforms: make labeling visible and preserve provenance information as much as possible.
- For brands: audit the agencies, influencers, and tools used in campaigns.
- For the public: check the source, date, context, and any labels before sharing.
The challenge will be to avoid two extremes. A label that is too discreet protects no one. An avalanche of incomprehensible mentions risks tiring the public and trivializing the danger. Transparency must be simple, coherent, and proportionate: sufficiently visible to inform, without leading to the belief that all computer-assisted content is necessarily false.
The Signal the World Can No Longer Ignore
Europe has not solved the global problem of deepfakes. However, it imposes a powerful idea: when a machine fabricates an appearance of reality, the public has the right to know. This rule can influence technical standards, advertising contracts, and editorial policies far beyond the borders of the Union.
The real test will begin after August 2. It will be essential to observe the quality of the labeling, the response of platforms, the checks by authorities, and any potential circumventions. But the course is set. In the attention economy, authenticity becomes a strategic advantage. Those who can prove the origin of their content will gain what artificial intelligence cannot produce alone: lasting trust.
Sources
- European Commission â Guidelines on Transparency Obligations of the AI Act, July 2026.
- AI Act Service Desk â Timeline and Obligations of Article 50, accessed July 31, 2026.
- Associated Press â Update on Newsroom Standards for Artificial Intelligence, July 23, 2026.
- IPTC â C2PA Guide and Content Credentials for Media, July 8, 2026.
