EU AI Act Article 50 Transparency Requirements: Complete Guide

1. The four Article 50 obligations

Article 50 creates four distinct transparency obligations. Each covers a different type of AI interaction:

2. Chatbot and conversational AI disclosure

Article 50(1) requires providers and deployers of AI systems designed to interact directly with natural persons to ensure those persons are informed that they are interacting with an AI system — unless this is obvious from context.

The disclosure must happen at the beginning of the interaction. A user clicking "accept" on a terms of service page that mentions AI somewhere is not sufficient.

Who must disclose

Both the provider (who built the chatbot platform) and the deployer (who deployed it on their website or in their product) share this obligation. In practice, the deployer is the party in direct contact with the user, so the disclosure typically falls to them to implement at the point of interaction.

What must be disclosed

The regulation does not prescribe exact wording, but the disclosure must be clear and intelligible to the user. It must convey that they are interacting with an AI system — not a human. The disclosure must be:

• Given before or at the start of the interaction
• In a form that the user can actually understand (not hidden in small print)
• In the language the user is interacting in

For chatbots that name themselves (common in customer service), adding "I'm an AI" or "I'm [Name], your AI assistant" to the opening message is generally sufficient. Use the free disclosure generator to produce compliant text in English, French, German, Dutch, or Spanish.

3. AI-generated content labelling

Article 50(2) requires providers of AI systems that generate synthetic content — text, images, audio, video — to ensure the output is labelled as machine-generated in a way that is machine-detectable. This means watermarking or metadata tagging, not just a visible label.

The obligation applies to content "intended to inform, entertain, or persuade natural persons." This is broad. It covers:

• AI-generated marketing copy published on a website
• AI-generated product descriptions auto-published to an e-commerce store
• AI-generated news summaries or social media posts
• AI-generated images used in advertising
• AI-generated audio for customer-facing voice systems

The machine-detectable marking requirement means content generation tools like ChatGPT, Jasper AI, Google Gemini, and Adobe Firefly must embed detectable signals in their output. As a deployer, you should verify that your content generation vendors comply with this — it is primarily a provider obligation, but you benefit from working only with providers who are building compliant watermarking systems.

4. Emotion recognition and biometric categorisation

Article 50(3) requires deployers of emotion recognition systems or biometric categorisation systems to inform the natural persons exposed to those systems that they are being subject to emotion recognition or biometric categorisation. This disclosure must be given in advance.

Emotion recognition AI that infers the emotional state of employees or customers — whether through facial analysis, voice tone, or physiological signals — falls under this obligation. Note that using emotion recognition in workplaces or educational institutions may also trigger the prohibited practice provisions in Article 5(1)(f) — separate from the transparency requirement and far more severe.

5. Deepfakes and synthetic media

Article 50(4) requires any person deploying an AI system to generate or manipulate image, audio, or video content that constitutes a deepfake to disclose that the content has been artificially generated or manipulated. The disclosure must be clearly visible or audible and appear throughout the content.

Exceptions apply for:

• Legitimate creative or artistic works where the AI-generated nature is clear from context
• Satire and parody where disclosure would impair the artistic expression and the AI-generated nature is apparent

The exception is narrow. Commercial use of synthetic media — advertising, political content, corporate video — does not qualify for creative exception. Disclose clearly.

6. Exceptions

Article 50(1) includes an explicit exception for AI systems authorised by law for the purpose of preventing, investigating, detecting, or prosecuting criminal offences, subject to appropriate safeguards. This covers specific law enforcement uses — it does not apply to commercial operators.

Article 50(1) also contains an implicit exception where the AI nature of the system is "obvious from context." A robot visibly moving and speaking to a visitor in a showroom is obviously AI. A customer service chatbot on a website named "Emma" that responds in natural language is not obvious — the disclosure obligation applies.

7. Compliant disclosure examples

Customer service chatbot

AI sales assistant

AI-generated article

Synthetic video

8. Affected tools

The following categories of tools trigger Article 50 obligations when deployed:

FAQ

Does Article 50 apply to internal Slack bots or internal AI tools?

Article 50(1) applies to AI systems "designed to interact directly with natural persons." An internal Slack bot used by employees who know it is AI is unlikely to trigger the obligation — the AI nature is obvious from context. However, if the tool could plausibly be mistaken for a human colleague, the disclosure is recommended as a matter of good practice and risk management.

We use a chatbot platform. Is it our obligation or the platform provider's?

Both. The chatbot platform provider must ensure their system has the capability to deliver compliant disclosures (Art. 50(1) obligation on providers). You, as the deployer, must ensure that disclosures are actually configured and displayed to users. Check that your chatbot platform's default opening message identifies the AI nature — and customise it if it does not.

Does Article 50 apply to email campaigns written with AI assistance?

If a human writes and reviews the email using AI assistance (drafting, spell check), the email is not purely AI-generated. Article 50(2) applies to automated generation pipelines where content goes from AI to publication without meaningful human review. If your marketing team edits AI drafts before sending, you are not in scope for the labelling obligation.

What is the fine for non-compliance with Article 50?

Non-compliance with Article 50 transparency obligations can result in fines of up to €15 million or 3% of worldwide annual turnover, whichever is higher. These are the same fines that apply to violation of high-risk deployer obligations under Article 26.