Transparency rules under Article 50 of the EU AI Act apply from August 2026. Some obligations for high-risk systems were postponed. For a business, the first step is to identify the AI system and its own role. Using AI to draft ordinary copy does not automatically mean every output must be labelled.
The European Commission timeline sets out the dates. Specific duties depend on the system and its use; this is a practical overview rather than legal advice for a particular deployment.
What already applies
From February 2, 2025: prohibitions of certain unacceptable practices and measures to support AI literacy among staff of AI system providers and deployers.
From August 2, 2025: rules for providers of general-purpose AI models.
From August 2, 2026: transparency rules under Article 50 and enforcement of applicable rules.
From December 2, 2027: rules for high-risk systems in Annex III, including certain recruitment, education and credit uses. For systems embedded in regulated products under Annex I, the date is August 2, 2028.
Chatbots and direct interaction
A provider of a system that directly interacts with people must ensure they are informed from the start that they are interacting with AI, unless this is obvious. A business deploying a customer-facing AI agent should check how its particular solution provides that notice. “I am an AI assistant” is a simple practical disclosure. The Commission's Article 50 FAQ explains the scope and exceptions.
Deepfakes, public-interest text and output marking
A deployer publishing content that meets the definition of a deepfake must disclose that it was artificially generated or manipulated. Deployers must also inform people exposed to emotion recognition or biometric categorisation. For AI-generated text, a disclosure rule concerns publication on matters of public interest; an exception applies where there is human review or editorial responsibility. There is no blanket duty to label every marketing text or every AI image. Realistic AI ads should be assessed in context.
A separate duty applies to providers of generative AI systems: outputs must be machine-readable as AI-generated or manipulated. The transition until December 2, 2026 only concerns certain systems placed on the market before August 2, 2026 and only this duty under Article 50(2). It is not a general deferral of deepfake disclosure. See the Commission's transparency guidance for the division of responsibilities.
A practical business checklist
1. Map AI use. Include customer chatbots, content creation and decisions affecting people.
2. Identify your role. For each system, distinguish offering it under your own name from using it as a deployer. The obligations differ.
3. Check communication and content. Review chatbot notices, potential deepfakes and public-interest text published without human editorial review.
4. Support staff AI literacy. Focus on the actual tools, output checking and data handling.
5. Get specialist advice for high-risk uses. Recruitment, credit assessment and similar areas may bring further duties; the postponement is temporary.
Penalties and Czech oversight
For certain Article 50 infringements, the AI Act provides a ceiling of €15 million or 3% of worldwide annual turnover; for SMEs the lower applicable ceiling applies. Any actual penalty depends on the infringement and proportionality. The Czech Telecommunication Office announced a dedicated AI unit from September 2026; its future powers depend on Czech legislation.
Frequently asked questions
An AI audit can identify systems, risks and practical next steps. For legally complex uses, I recommend specialist legal advice.
Free consultationI build AI solutions for businesses — from chatbots to knowledge systems. I write about what actually works, no buzzwords. More about me →