← Guides

Klarframe · Guides

EU AI Act Article 50: Transparency Obligations in 2026

By Oliver Condurache, Co-founder, Klarframe · Updated

Article 50 of the EU AI Act (Regulation (EU) 2024/1689) has applied since 2 August 2026, and the Digital Omnibus did not postpone it. Chatbots, voice agents and AI avatars must tell people they are interacting with AI unless this is obvious. Deployers must label deepfakes, inform people about emotion recognition and disclose AI-generated text on matters of public interest when no one holds editorial responsibility. Providers of generative AI must mark outputs in a machine-readable way. Only this Article 50(2) marking duty has a transition until 2 December 2026, and only for systems placed on the market before 2 August 2026.

What does Article 50 of the EU AI Act regulate?

Article 50 of the EU AI Act defines when people must be told that they are dealing with artificial intelligence. The provision is part of Regulation (EU) 2024/1689, the European AI Act, and has applied since 2 August 2026. Unlike the rules for high-risk AI, Article 50 covers many everyday applications: website chatbots, phone assistants, digital avatars, image and video generators, and AI-generated text.

The idea is simple: anyone talking to a machine or seeing artificially generated content should be able to recognise it. Article 50 contains no technology ban. It requires openness. Depending on the case, the obligations fall on providers, meaning companies that develop an AI system or place it on the market under their own name, or on deployers, meaning companies that use an AI system under their own authority.

Under Article 50, the information must be given at the latest at the time of the first interaction or exposure, in a clear and distinguishable manner, and must take accessibility requirements into account.

Which obligations apply to providers and deployers?

Article 50 contains several obligations that concern different roles and applications. For companies it helps to map each of their own AI applications to one of these categories:

  • Interaction with people (providers, Article 50(1)): AI systems that interact directly with people must be designed so that those people are informed they are interacting with AI, unless this is obvious to a reasonably well-informed person.
  • Synthetic content (providers of generative AI, Article 50(2)): AI-generated audio, image, video or text must be marked in a machine-readable format and be detectable as artificially generated or manipulated. There are exceptions, for example for purely assistive editing functions that do not substantially alter the input.
  • Emotion recognition and biometric categorisation (deployers, Article 50(3)): the people exposed must be informed about the operation of such a system.
  • Deepfakes (deployers, Article 50(4)): anyone generating or manipulating image, audio or video content that resembles real people, places or events must disclose that it was artificially generated or manipulated.
  • AI-generated text (deployers, Article 50(4)): text published to inform the public on matters of public interest must be disclosed as AI-generated, unless it has undergone human review or editorial control and a person or company holds editorial responsibility.

What does Article 50 mean for chatbots, voice agents and AI avatars?

For chatbots, voice agents and interactive AI avatars, the key obligation is disclosing the AI interaction. It has applied since 2 August 2026 without any grace period. A website chatbot should make clear that an AI is answering, for example in the chat window and in the first message. A voice agent on the phone should introduce itself as an AI assistant at the start of the call, because callers can rarely tell whether a person or a machine is speaking. The Commission guidelines read the exception for obvious cases narrowly.

With AI avatars two aspects come together. A video avatar that talks to visitors and answers questions interacts with people and must be recognisable as AI. If the avatar also looks like a real, existing person, it may count as a deepfake, which makes labelling it as artificially generated all the more important. For evidently artistic, satirical or fictional works, Article 50 provides a lighter disclosure duty that should not hamper the display of the work.

In practice a clear sentence at the start works well, such as “I am an AI assistant for …”, supported by a visible notice in the interface and a transparency page explaining where AI is used and how people can reach a human.

When must deepfakes and AI-generated text be labelled?

Under Article 50, deepfakes must be disclosed whenever AI-generated or AI-manipulated image, audio or video content resembles real people, objects, places or events and could falsely appear authentic. This includes realistic promotional videos with synthetic presenters, cloned voices or photos that recreate real scenes. Purely abstract graphics or clearly stylised illustrations usually do not fall under the deepfake definition.

For AI-generated text the obligation is narrower. It covers text published to inform the public on matters of public interest, such as news or articles on societal topics. If such text has been reviewed by people and a person or company holds editorial responsibility, the labelling duty does not apply. Anyone relying on this exception should keep brief records of the human review. Product copy or internal documents are usually not covered, although voluntary transparency can still make sense.

Separately from deployers’ duties, providers of generative AI systems must ensure their outputs are marked in a machine-readable way and are detectable, for example through watermarks, metadata or cryptographic methods. Companies using off-the-shelf tools should therefore check whether their provider supplies this marking and whether it survives their own further processing.

What is the current status of guidelines, the code of practice and deadlines?

To our knowledge as of October 2026, the timeline after the Digital Omnibus on AI (Regulation (EU) 2026/1744) is as follows:

  • 2 February 2025: Article 4 (AI literacy) and the prohibitions in Article 5 apply.
  • 2 August 2025: the penalty rules and the obligations for general-purpose AI models apply.
  • 10 June 2026: the Commission publishes the final Code of Practice on Transparency of AI-Generated Content, covering marking (providers) and labelling (deployers). In July 2026 the Commission and the AI Board confirmed it as an adequate means of demonstrating compliance. Signing it is voluntary.
  • 16 and 29 June 2026: the European Parliament and the Council approve the Digital Omnibus on AI.
  • 20 July 2026: the Commission publishes its final, non-binding guidelines on Article 50. They shape how supervisory authorities interpret the obligations.
  • 24 July 2026: Regulation (EU) 2026/1744 is published in the Official Journal of the EU and enters into force on 27 July 2026.
  • Since 2 August 2026: Article 50 applies. Disclose AI interaction, label deepfakes and AI-generated text, inform people about emotion recognition. Generative AI systems placed on the market from this date must also meet the machine-readable marking duty of Article 50(2) immediately.
  • From 2 December 2026: the transition for Article 50(2) marking ends for generative AI systems placed on the market before 2 August 2026. On the same day, new Article 5 prohibitions take effect for AI-generated non-consensual intimate content and child sexual abuse material.
  • Postponed: obligations for high-risk AI under Annex III apply from 2 December 2027, and for products under Annex I from 2 August 2028. Article 50 is not affected by this.

What fines apply for breaches of Article 50?

Under Article 99, breaches of the transparency obligations in Article 50 can lead to fines of up to EUR 15 million or up to 3 % of total worldwide annual turnover for the preceding financial year, whichever is higher. For SMEs and start-ups, the lower of the two amounts applies. Enforcement lies with the national market surveillance authorities.

Guidelines, codes and details may be adjusted further. Check the current status directly with the European Commission and in the Official Journal of the EU before making decisions.

What should companies do now, and how does Klarframe implement Article 50?

For most companies, Article 50 can be tackled with a manageable review. These steps help to get oriented:

  • Create an inventory: list every AI application that interacts with people or generates content, such as a chatbot, phone assistant, avatar, image and video generation or text.
  • Clarify the role: record whether the company is a provider, a deployer or both. This determines which obligations apply.
  • Design notices: disclose the AI interaction at the start, in clear and accessible wording.
  • Review content: label realistic AI images, voices and videos, and preserve the markings from your tools.
  • Plan for existing systems: if you provide generative AI that was on the market before 2 August 2026, implement machine-readable marking by 2 December 2026.
  • Offer human contact: name a clear route to a person, for example by email or contact form.
  • Document: record decisions and notices, and update them when new guidance appears.

What does Article 4 require on AI literacy?

Article 4 of the EU AI Act has applied since 2 February 2025 and concerns all providers and deployers of AI systems, regardless of risk level. The Digital Omnibus softened the duty: instead of ensuring a sufficient level of AI literacy, companies must now take measures to promote the AI literacy of their staff and of other persons using AI systems on their behalf. The Commission and the Member States support this, in particular for SMEs.

In practice the duty remains. Short training on the tools in use, clear internal rules and documentation of these measures are sensible. To identify gaps in AI literacy, processes and transparency in a structured way, you can use the free Klarframe AI Readiness Check as a self-assessment.

Klarframe discloses AI interactions as required by the EU AI Act. The AI assistant Emma introduces herself as AI, realistic AI-generated or AI-manipulated image, audio and video content is labelled directly next to the content, and anyone who prefers a person can reach the team by email or contact form. The details are on the Klarframe transparency page at https://klarframe.com/en/ki-transparenz. In client projects such as voice agents and video avatars, the AI notice is part of the conversation and interface design.

Not legal advice: this guide summarises Article 50 in general terms and to the best of our knowledge as of October 2026. It does not replace individual legal advice or a review of your specific case. Whether and how the obligations apply to your company should be clarified with a lawyer or your data protection advisor.

Frequently asked questions

When does Article 50 of the EU AI Act apply?

Article 50 of the EU AI Act has applied since 2 August 2026. Since then, companies must among other things disclose when people interact with AI and label deepfakes. The Digital Omnibus (Regulation (EU) 2026/1744) introduced a transition until 2 December 2026 only for machine-readable marking under Article 50(2). It covers generative AI systems placed on the market before 2 August 2026. Newer systems must mark their outputs immediately.

Has the Digital Omnibus postponed Article 50?

No. The Digital Omnibus on AI, in force since 27 July 2026, left Article 50 applying from 2 August 2026 as a rule. The main postponements concern high-risk AI: Annex III obligations move to 2 December 2027 and Annex I obligations to 2 August 2028. For Article 50 there is only a transition until 2 December 2026 for machine-readable marking by generative AI systems placed on the market before 2 August 2026.

What fines apply for breaches of Article 50?

Under Article 99 of the EU AI Act, breaches of the transparency obligations in Article 50 can be fined up to EUR 15 million or up to 3 % of total worldwide annual turnover for the preceding financial year, whichever is higher. For SMEs and start-ups, the lower of the two amounts applies. National market surveillance authorities enforce these rules.

Does a website chatbot have to be labelled as AI?

Yes, as a rule. Article 50 requires that people are informed when they interact with an AI system, unless this is obvious. A website chatbot should therefore be clearly recognisable as AI by the first message at the latest, for example through a notice in the chat window. This duty has applied since 2 August 2026 without a grace period. A route to a human and a transparency page are also advisable.

Do I have to label AI-generated blog posts?

The labelling duty for AI-generated text under Article 50 covers text intended to inform the public on matters of public interest. It does not apply if the text was reviewed by people or under editorial control and someone holds editorial responsibility. For typical company blogs with human editing, the duty therefore often does not apply, though a case-by-case check remains sensible.

What is a deepfake under the EU AI Act?

A deepfake under the EU AI Act is AI-generated or manipulated image, audio or video content that resembles existing persons, objects, places, entities or events and would falsely appear to be authentic or truthful. Anyone deploying such content must disclose under Article 50 that it has been artificially generated or manipulated. A lighter duty applies to evidently artistic or satirical works.

How does Klarframe implement the transparency obligations?

Klarframe discloses AI interactions as required by the EU AI Act. The AI assistant Emma introduces herself as AI, realistic AI-generated image, audio and video content is labelled next to the content, and a person can be reached by email or contact form. Details are on the transparency page at https://klarframe.com/en/ki-transparenz. In client projects the AI notice is part of the conversation design.

  • Video Avatars →

    Digital video avatars welcome visitors, explain products and guide customers through content and onboarding.

  • Inbound Voice Agent →

    An inbound voice agent answers naturally, captures information and routes each request reliably.

  • AI Meeting Concierge →

    An AI assistant for intelligent scheduling and professional customer care. It saves time, handles research and prioritisation and makes room for more qualified customer conversations.

Klarframe AI Readiness Check

How ready is your company for AI?

About ten minutes, instant result with your stage, a 90-day plan, suitable solutions and notes on the EU AI Act.

Start the check

Want to know whether this fits your business? Your regional Klarframe partner will advise you personally.

Book a conversation →