EU AI Act Article 50: Transparency & AI Disclosure Requirements
Last updated 2 August 2026 — reflects the Digital Omnibus on AI.
Article 50 is the EU AI Act obligation that touches the most companies, and it is live from August 2026. Here is exactly what transparency, watermarking, and deepfake disclosure require — and how to comply.
On this page
Article 50 of the EU AI Act is the part that affects the most companies, and as of 2 August 2026 it is live. It sets the transparency obligations for "limited risk" AI — chatbots, generative AI output, deepfakes, and emotion-recognition systems. If your product talks to users or generates content, this is the article you must comply with now.
The four transparency obligations
1. AI system interaction (chatbots)
Providers must ensure that people are informed they are interacting with an AI system, unless it is obvious from the context to a reasonably well-informed person. In practice: a visible disclosure on chatbots and AI voice agents.
2. Marking of AI-generated content
Providers of generative AI must mark synthetic audio, image, video, and text outputs in a machine-readable format so they can be detected as artificially generated — for example via watermarking or content credentials (C2PA-style metadata).
3. Deepfake disclosure
Deployers who create or manipulate image, audio, or video content that resembles real people, objects, or events (a "deepfake") must disclose that the content is artificially generated or manipulated.
4. AI-generated text on matters of public interest
Where AI-generated text is published to inform the public on matters of public interest, that must be disclosed — with a carve-out where the content underwent human review and a person or organisation holds editorial responsibility.
A related duty applies to emotion recognition and biometric categorisation: deployers must inform the people exposed to those systems.
Watermarking and machine-readable marking
The marking obligation is the most technical. The content must be detectable by machines, robust, and — as far as technically feasible — interoperable. Two practical approaches dominate:
- Watermarking — embedding a signal directly in the generated media.
- Content credentials / C2PA — attaching cryptographically signed provenance metadata to the file.
Systems already on the market on 2 August 2026 have a grace period until 2 December 2026 to implement machine-readable marking. New systems should build it in from the start.
Who has to do what: provider vs deployer
| Obligation | Falls on |
|---|---|
| Inform users they are interacting with AI | Provider |
| Machine-readable marking of AI-generated output | Provider |
| Disclose deepfakes | Deployer |
| Disclose AI text on matters of public interest | Deployer |
| Inform people subject to emotion recognition / biometric categorisation | Deployer |
Many companies are both provider and deployer, so most teams will need to cover several of these.
What non-compliance costs
Breaches of Article 50 sit in the AI Act's second penalty tier: up to €15 million or 3% of global annual turnover, whichever is higher. It is not the top tier reserved for prohibited practices, but it is more than enough to matter.
Article 50 compliance checklist
- Add a clear AI disclosure to every chatbot and voice agent.
- Turn on machine-readable marking (watermark or content credentials) for any generative output — before 2 December 2026 if the system is already live.
- Label deepfakes and AI-manipulated media wherever you publish them.
- Disclose AI-generated public-interest text, or document your human-editorial-review process to rely on the carve-out.
- Notify users exposed to emotion recognition or biometric categorisation.
- Keep records of how each disclosure is implemented — that is your evidence of compliance.
Shipping a chatbot or AI feature into the EU?
GenAI Labs builds EU Compliant Chat — conversational AI with Article 50 transparency, logging, and human-oversight controls built in, so you meet the AI Act's requirements without slowing your roadmap.
See EU Compliant Chat →Frequently asked questions
What does Article 50 of the EU AI Act require?
Article 50 sets transparency obligations for limited-risk AI. Providers must tell users when they are interacting with an AI system and mark AI-generated content in a machine-readable way. Deployers must disclose deepfakes, disclose AI-generated text published on matters of public interest (unless it had human editorial review), and inform people subject to emotion recognition or biometric categorisation.
When do the EU AI Act transparency rules apply?
Article 50 transparency obligations apply from 2 August 2026. Systems already on the market have a grace period until 2 December 2026 to implement machine-readable marking of AI-generated content.
How do I mark AI-generated content to comply with the AI Act?
The Act requires AI-generated audio, image, video, and text to be marked in a machine-readable format that is detectable as artificial. The two common approaches are watermarking (embedding a signal in the media) and content credentials / C2PA (attaching cryptographically signed provenance metadata). The marking should be robust and, where feasible, interoperable.
Does my chatbot need to disclose that it is AI under the EU AI Act?
Yes. Under Article 50, providers must ensure users are informed they are interacting with an AI system, unless that is obvious from the context. A visible disclosure on your chatbot or AI voice agent satisfies this. This obligation has applied since 2 August 2026.
What are the penalties for breaching Article 50?
Breaches of Article 50 fall in the second penalty tier of the EU AI Act: up to €15 million or 3% of global annual turnover, whichever is higher.