Frequently Asked Questions
Answers to the most common questions about our initiative
No. This proposal does not seek to ban or limit generative AI. It only asks that AI-generated content be identifiable as such. This is a question of transparency, not a restriction on creation.
Labelling does not prevent parody or artistic creation. A satirical piece labelled "AI-generated" remains perfectly legal and shareable. If anything, this strengthens protections for artists by clearly distinguishing generated works from authentic recordings.
This is precisely why the metadata is cryptographically signed. Any modification to the file invalidates the signature, making manipulation detectable. Platforms would also be prohibited from performing re-encoding that strips this metadata.
The proposal includes an obligation for platforms to detect and reject re-uploads of AI content without a marker, where that content has previously been identified as AI-generated in their system.
AI generation tools (software, APIs, platforms such as Sora, Midjourney, Stable Diffusion, etc.) would be legally required to embed this marking automatically at output. End users would not be expected to do this manually.
Yes. The Coalition for Content Provenance and Authenticity (C2PA) is a provenance standard voluntarily adopted by Adobe, Microsoft, Google and Sony. It can serve as a working basis. But a European legal obligation cannot rely exclusively on a standard governed by private actors — hence the need for an open, auditable and sovereign framework.
The proposal primarily targets audio-visual content (video, audio), which carries the highest disinformation risk. Its extension to still images and AI-generated text could be addressed in a subsequent phase.