This article is for information only and isn't a substitute for legal advice. Rules vary by jurisdiction — for your specific situation, check with a legal professional.
What Article 50 actually says
Starting August 2, 2026, the EU AI Act imposes a transparency obligation on content produced by artificial intelligence. In practice: providers of generative AI systems must ensure their outputs are marked in a machine-readable format and detectable as artificially generated or manipulated — and the businesses that use these systems (the “deployers”) must clearly disclose any deepfake, meaning any realistic image, audio or video content depicting a person, object, place or event, when it has been generated or substantially altered by AI.
Who is affected
Any business or independent creator who generates visuals, audio or video with AI, or who makes a generative AI tool available to their team or clients, falls within the scope of this obligation — not just the large model providers. A freelancer who delivers an AI avatar video to a client, or an agency that publishes AI-generated visuals on a brand's social accounts, are covered in the same way as a large corporation.
A two-stage timeline
- August 2, 2026 — the disclosure obligations take effect: a clear notice for the end user, deepfake disclosure, and information about content intended to inform the public.
- December 2, 2026 — end of the technical grace period for machine-readable marking (metadata, digital watermark embedded in the file).
Key takeaway. The two obligations don't replace each other: an invisible technical mark inside the file on one side, and an understandable notice for the person watching or listening on the other.
What a non-compliant business risks
Failing to meet Article 50's transparency obligations can trigger a fine of up to €15 million or 3% of worldwide annual turnover, whichever is higher — a scale comparable to GDPR penalties, and one that applies to small businesses just as much as to large corporations.
What this changes for image, video and voice
This obligation directly touches the three most common uses of creative AI:
- Image — any photorealistic image depicting a person or event needs to be disclosed as generated. Full detail in our AI image generation guide.
- Video — a generated clip shared publicly falls under the same rule, especially if it looks realistic. See our creating a video with AI guide.
- Avatar and voice — this is the case most directly targeted by the notion of a deepfake: a realistic avatar or voice clone must be disclosed unambiguously. Our AI avatar guide already recommended this default transparency before the law made it mandatory.
Frequently asked questions
Article 50 targets providers of generative AI systems and deployers who use AI in a professional context — which includes a freelancer or agency publishing AI-generated content for a client or for their own business communication. When in doubt, disclosing is the safer default.
Both obligations exist side by side and neither replaces the other: a machine-readable technical mark embedded in the file (metadata, digital watermark) on one hand, and a clear, understandable disclosure for the person viewing the content on the other — especially for any deepfake (a realistic image, audio or video depicting a person, object or event).
Failing to meet Article 50's transparency obligations can trigger a fine of up to €15 million or 3% of worldwide annual turnover, whichever is higher — roughly the same order of magnitude as GDPR penalties.
No: a technical grace period runs until December 2, 2026, specifically for machine-readable marking (metadata, watermarking). The clear public-disclosure obligations (visible notice, deepfake disclosure), however, are expected to apply from August 2, 2026.
Go further
- AI avatar: creating your digital twin
What an AI avatar can actually do today (training, LinkedIn, sales), its limits, and how to create one without filming anything.
- How to translate a video into another language with an AI voice?
The complete method for adapting a video into several languages: extracting the script, translating it, generating the voice, and re-syncing lip movement if needed.
- AI voice or human voiceover: what's the difference for a professional video?
Where AI wins by a wide margin (cost, turnaround, revisions), where human voiceover keeps a real edge, and how to decide based on your project rather than on principle.
- Free or paid AI video generator: which one should you start with?
What free plans actually limit (watermark, duration, quota), and at what point a subscription becomes more cost-effective than stacking free accounts.
See exactly what ToolAcces includes — credited API engines and mutualized access, in a single subscription starting at €39.99/month, no commitment. View pricing.