Article 50 of the EU AI Act introduces specific transparency obligations for companies that build or deploy certain types of AI systems. Unlike high-risk requirements that demand extensive documentation and conformity assessments, Article 50 focuses on a simpler but equally important principle: people should know when they are interacting with AI.
These obligations apply from August 2025, making them among the earliest enforcement deadlines in the AI Act timeline. For many SMEs and SaaS companies, Article 50 may be the first obligation that actually requires product changes.
What does Article 50 require?
Article 50 sets out transparency rules for several specific categories of AI systems. The requirements differ depending on what the system does and who operates it.
1. AI systems that interact directly with people
If your product includes a chatbot, virtual assistant, AI-powered support agent, or any system that communicates with users, those users must be informed that they are interacting with AI — unless this is obvious from the circumstances.
This applies to providers and deployers. In practice, it means adding a visible label, disclaimer, or UI indicator so users understand the nature of the interaction.
2. AI-generated or AI-manipulated content
If your system generates synthetic text, images, audio, or video — including deepfakes — the output must be marked as AI-generated. This applies to:
- AI-generated images and videos (e.g. marketing visuals, avatars)
- Synthetic voice or audio content
- Deepfake-style content that depicts real people or events
- AI-generated text presented as factual content (e.g. articles, summaries)
The marking must be done in a machine-readable format where technically feasible, and must be clearly communicated to users.
3. Emotion recognition and biometric categorisation
If your system performs emotion recognition or biometric categorisation, people who are subject to it must be informed. This is relevant for companies using AI to analyse facial expressions, voice tone, or physiological signals.
Who does Article 50 apply to?
Article 50 applies to both providers (companies that develop or place AI systems on the market) and deployers (companies that use AI systems in their operations or products).
For SaaS companies, this is especially important. If you embed AI features into a product used by customers, you may be considered a provider. If your customers deploy your AI features to their own end users, both of you may have transparency obligations. Read more about how this affects SaaS teams in our SaaS guide.
How is Article 50 different from high-risk obligations?
High-risk AI systems (covered under Articles 6–49) require conformity assessments, technical documentation, risk management systems, and human oversight. Article 50 is lighter — it focuses on disclosure rather than governance infrastructure.
However, a single AI system can trigger both Article 50 transparency obligations and high-risk obligations. For example, an AI hiring tool that interacts with candidates may need transparency labelling (Article 50) and full high-risk compliance (Annex III). Use the free AI Act scanner to understand which obligations may apply to your systems.
Practical steps to prepare
Preparing for Article 50 is straightforward compared to other parts of the AI Act, but it does require deliberate product and communication decisions.
- Audit user-facing AI. Identify every touchpoint where users interact with AI directly — chatbots, copilots, recommendation engines, AI-generated content.
- Add transparency indicators. Use clear labels such as "AI-generated", "Powered by AI", or "You are chatting with an AI assistant." The key is that users should not be misled.
- Mark AI-generated content. If your platform produces synthetic text, images, or media, implement machine-readable watermarks or metadata tags where feasible.
- Document your approach. Keep a simple record of what transparency measures you have applied and why. This supports accountability if questions arise.
- Review third-party AI. If you use APIs or models from third parties, check whether the outputs need transparency labelling on your side.
For a broader compliance framework, see our AI Act compliance checklist.
Common questions about Article 50
Does Article 50 apply to internal tools?
If an AI system only interacts with your own employees (e.g. an internal copilot), transparency obligations are lighter. However, if the system performs emotion recognition or biometric categorisation on employees, they must still be informed.
What if the AI interaction is obvious?
The regulation includes an exception: if it is "obvious to a reasonably well-informed person" that they are dealing with AI, explicit disclosure may not be required. In practice, it is safer to disclose.
What happens if we don't comply?
Non-compliance with transparency obligations can result in fines of up to €15 million or 3% of global annual turnover, whichever is higher. For SMEs, the Digital Omnibus may introduce proportionate enforcement, but the obligations themselves still apply.
Timeline
Article 50 transparency obligations apply from 2 August 2025. This is earlier than the high-risk system requirements (August 2026). Companies should prioritise Article 50 compliance now. See the full AI Act timeline for all key dates.
Final takeaway
Article 50 is one of the most actionable parts of the EU AI Act. It does not require complex governance frameworks or conformity assessments. It requires honesty: tell people when they are interacting with AI, and label content that AI has created. For most companies, this is a product design decision, not a legal one.
Start with our EU AI Act checklist for a high-level overview. Then run a free AI Act scan to identify which of your systems may trigger Article 50 obligations. Use the AI Act compliance checklist for detailed next steps, or read the EU AI Act Guide for full context. Return to the ActNavigator homepage for all tools.