EU AI Act Article 50 has applied since 2 August 2026. Chatbots and other AI systems that talk to people must tell them they are dealing with AI. Generative AI output must carry a machine-readable mark, and deployers must label deep fakes and some published text. Only the marking duty, for systems already on the market, waits until 2 December 2026.

Picture an online retailer based in the EU. It has a support chatbot on its website, an image tool in its app that lets shoppers try products in their own room, and a blog its marketing team drafts with AI. In September the team read that the AI Act had been delayed and parked the topic until 2027. The delay was real, but it was for high-risk AI systems. None of those three features is high-risk, and the rules that cover them are already in force.

This post is not legal advice; check your own situation with a lawyer.

What did the AI Act delay, and what did it leave alone?

The delay moved the high-risk rules, not the transparency rules. Regulation (EU) 2026/1744, known as the Digital Omnibus on AI, was adopted on 8 July 2026, published in the Official Journal on 24 July and entered into force on 27 July.

It pushed the main high-risk obligations to 2 December 2027 for systems listed in Annex III, such as those used in hiring or credit decisions, and to 2 August 2028 for AI built into products covered by Annex I. It did not change when Article 50 applies. Article 113 of the EU AI Act (Regulation (EU) 2024/1689) still sets the general date of application at 2 August 2026, and Article 50 falls under that date.

What are the four obligations in Article 50?

Article 50 sets four transparency duties, split between providers and deployers. A provider is the organisation that develops an AI system, or has it developed, and puts it on the market or into service under its own name. A deployer is an organisation that uses an AI system under its authority in a professional setting.

• 50(1): design systems that interact with people so they know they are dealing with AI. Who: Provider. Applies from: 2 August 2026.

• 50(2): mark synthetic audio, image, video and text in a machine-readable, detectable way. Who: Provider of the generative system. Applies from: 2 August 2026; 2 December 2026 for systems placed on the market before 2 August 2026.

• 50(3): inform people exposed to emotion recognition or biometric categorisation. Who: Deployer. Applies from: 2 August 2026.

• 50(4): disclose deep fakes, and AI-generated text published to inform the public on matters of public interest. Who: Deployer. Applies from: 2 August 2026.

Article 50(5) adds how: the information must be clear and distinguishable, given at the latest at the first interaction or exposure, and meet accessibility requirements.

The Commission's guidelines on Article 50, published on 20 July 2026, explain each duty with examples. They are non-binding; only the Court of Justice of the EU can give an authoritative interpretation.

Does your customer-service chatbot need an AI disclosure?

In most cases, yes, and it has needed one since 2 August 2026. The guidelines list "AI chatbots embedded in online platforms or assistance support tools (helpdesks)" as a case where the duty applies.

The duty sits with the provider. That is not only the model vendor. The guidelines give the example of a company that develops a chatbot in-house and puts it into service under its own name: that company is the provider. A company that modifies an existing generative system and offers it under its own name also becomes a provider of the new system.

When the "obvious" exception applies

Article 50(1) does not apply where it is obvious to a reasonably well-informed person that they are talking to AI. The guidelines say this exception should be read restrictively, and that general public awareness of chatbots is not enough. Their examples of obvious cases are narrow: a code-review assistant used only by professional developers, or an internal assistant for trained staff in HR, legal or IT support. A public-facing chatbot that customers could mistake for a person does not qualify.

What counts as a disclosure

The guidelines suggest a first-turn greeting that says the chatbot is AI, a clear label or banner placed close to the interaction, and, for voice, a spoken statement at the start of the call. They also list what is not enough on its own:

• a disclosure only in the terms and conditions or documentation

• a machine-readable mark the user cannot see

• a vague name like "assistant", or a human photo on the bot

• a site-wide line such as "Services on this website use AI"

What does the 2 December 2026 grace period cover?

It covers one obligation for one group: providers of generative AI systems placed on the market before 2 August 2026 have until 2 December 2026 to meet the Article 50(2) marking duty. The Digital Omnibus added this as Article 111(4) of the AI Act. Recital 38 of the Omnibus describes it as a four-month transitional period.

What the grace period does not cover

• Chatbot disclosure. The guidelines say systems that are partly interactive and partly generative get the extra time only for marking. Telling people they are talking to AI was due on 2 August 2026.

• Deployer duties. Labelling deep fakes and public-interest text has no grace period.

• New systems. A generative feature launched on or after 2 August 2026 must be marked from day one.

What marking means

Marking is a machine-readable signal in the output that lets software detect it was AI-generated. Recital 133 of the AI Act names watermarks, metadata, cryptographic provenance methods, logging and fingerprints as examples. The guidelines say solutions should be effective, interoperable, robust and reliable as far as technically feasible. If your app generates images or audio with a vendor's model, ask the vendor which marks survive your own processing, such as resizing or compression.

Do you have to label AI-generated marketing content and blog posts?

Only some of it. The labelling duty for deployers covers deep fakes and text "published with the purpose of informing the public on matters of public interest". Most marketing copy falls outside that.

Deep fakes

A deep fake is AI-generated or manipulated image, audio or video that resembles real people, objects, places or events and would falsely appear authentic. If you publish one, disclose it. For clearly artistic, satirical or fictional work, the disclosure can be limited so it does not spoil the work.

Text on matters of public interest

The guidelines place ordinary advertising and product descriptions outside the scope, unless they make claims about health, consumer safety or sustainability. They place inside it, for example, an AI-manipulated lifestyle article on diets and a disease, and AI-manipulated corporate reports on a listed company's website.

The human review exception

Text needs no label if it has gone through human review or editorial control, and a person or company holds editorial responsibility for it. The guidelines set a real bar. Fact-checking is the minimum. Spell-checks, an editorial policy on paper, or a quick approval without engagement do not count. If AI rewrites the text after sign-off, the exception no longer applies. The guidelines also recommend publishing who holds editorial responsibility, for example in your site's legal information.

Content generated and published before 2 August 2026 does not need to be labelled retroactively. Text generated before that date but published after it does.

What should a business check before 2 December 2026?

Start with an inventory, because the duty depends on what each AI feature does and whose name is on it:

1. List every AI feature customers or the public can see: chat, voice, email agents, image or audio generation, and published text.

2. For each, record whether you are the provider (built or offered under your name), a deployer, or both.

3. For every interactive feature, confirm a visible AI disclosure at the first interaction. This was due on 2 August 2026.

4. For every generative feature you provide, confirm machine-readable marking. If it launched before 2 August 2026, the deadline is 2 December 2026; if later, it is already due.

5. Ask each model vendor in writing what marking they apply and whether it survives your pipeline.

6. Write down your editorial process for public-interest text: who reviews, what they check, and who holds editorial responsibility.

7. Check the disclosures meet accessibility requirements, for example screen-reader text and a spoken line on voice channels.

8. Decide whether to follow the Code of Practice on Transparency of AI-generated Content, published in final form on 10 June 2026. It is voluntary; if you don't follow it, you must show compliance by other adequate means.

The Commission's Article 50 FAQ says fines can reach 15 million euros or 3% of total worldwide turnover for the preceding financial year, and proportionality can be considered for SMEs and small mid-caps. National market surveillance authorities do most of the enforcement.

How we approach it at Vizio AI

We treat Article 50 as a design and engineering task as much as a legal one. With the client, we list every AI feature in production, trace which model and vendor sits behind it, and mark each one as provider, deployer or both. The client's lawyer confirms the reading; we don't replace that step.

Then we build the controls into the system rather than the policy document. Disclosure goes in the first message and the interface, not the footer. Marking is checked at the end of the pipeline, after every resize and re-encode. Editorial sign-off is recorded in the publishing workflow, so the human review exception can be shown, not just claimed.

This is part of the work of our DX Studio. If compliance is one reason an AI programme has stalled, Why AI Transformations Fail covers the other usual causes. If your chatbot or generator was built quickly with AI coding tools, Your AI-Built App Works in the Demo. Here's What Breaks With Real Users is the launch checklist to run alongside this one.

Sources

• EUR-Lex, Regulation (EU) 2024/1689 (Artificial Intelligence Act): https://eur-lex.europa.eu/eli/reg/2024/1689/oj/eng

• EUR-Lex, Regulation (EU) 2026/1744 (Digital Omnibus on AI): https://eur-lex.europa.eu/eli/reg/2026/1744/oj/eng

• AI Act Service Desk, Article 50: Transparency obligations for providers and deployers of certain AI systems: https://ai-act-service-desk.ec.europa.eu/en/ai-act/article-50

• AI Act Service Desk, Article 113: Entry into force and application: https://ai-act-service-desk.ec.europa.eu/en/ai-act/article-113

• European Commission, Transparency obligations under Article 50 of the AI Act (FAQ): https://digital-strategy.ec.europa.eu/en/faqs/transparency-obligations-under-article-50-ai-act

• European Commission, Commission publishes guidelines on transparency obligations for providers and deployers of certain AI systems: https://digital-strategy.ec.europa.eu/en/news/commission-publishes-guidelines-transparency-obligations-providers-and-deployers-certain-ai-systems

• European Commission, Guidelines on the implementation of the transparency obligations for certain AI systems under Article 50 of the AI Act, C(2026) 5054 final: https://ai-act-service-desk.ec.europa.eu/sites/default/files/2026-07/guidelines_on_the_implementation_of_the_transparency_obligations_for_certain_ai_systems_under_article_50_of_the_ai_act_bzptwqhk0ikg1dtlddap41psfy_131215.pdf

• European Commission, Code of Practice on Transparency of AI-generated Content: https://digital-strategy.ec.europa.eu/en/policies/code-practice-ai-generated-content