Skip to main content

Merchant radar

Merchant radar · E-commerce

AI transparency obligations: chatbots, AI images and deep fakes in online shops

Deadline
Status
applies since 2 August 2026
Date
confirmed
Affects
all online shops
Updated on

Since 2 August 2026, Article 50 of the AI Act has required that people be told when they deal with AI and that deep fakes be disclosed, by AI providers or by shops as deployers, depending on the case.

In short

  • What: Under Article 50, people must be told when they interact with an AI system; realistic AI images, video and audio must be disclosed as artificial.
  • Who: Providers that develop AI, or have it developed, and offer it under their own name; in some cases, shops using AI under their authority (deployers).
  • Which products: Examples in the Commission's guidelines include customer service chatbots, misleading AI product images and advertising videos with simulated people.
  • When: Since 2 August 2026. Providers of generative AI systems placed on the market before then have until 2 December 2026 to apply machine-readable marking.
  • What to check: Where AI in your shop talks to customers or creates images, and whether that needs a notice.

Does this apply to me?

Your duties depend on your role under the AI Act: provider or deployer.

Who has which duty
RoleDutyExample from the guidelines
Provider that develops a chatbot, or has one developed, and offers it under its own nameThe provider must design and build the chatbot so that people are told they are interacting with an AI system.According to the Commission's guidelines, a company that builds a chatbot in-house and uses it under its own name or trademark is a provider too. A company that offers a chatbot under its own name remains the provider, even when its customers integrate the chatbot into their own processes without changing it.
Provider of AI that generates text, images, audio or videoThe provider must mark the output in a machine-readable format.According to the Commission's guidelines, grammar and spelling correction, and translation, count as standard editing and need no marking. Short outputs such as image captions and alt text (the written description of an image) need no marking either. AI summaries do need marking, and so does rewriting that changes style, structure and meaning.
Shop that modifies another provider's generative AI system and uses the new system under its own nameAccording to the Commission's guidelines, the shop becomes the provider of the new system.The guidelines name new training data as one way of modifying a system. Training data is the data an AI system learns from.
Shop as deployer that uses AI to create or alter deep fakes (realistic but artificial images, audio or video of people, objects or places)The deployer must disclose that the content has been artificially generated or manipulated.According to the Commission's guidelines, objects include consumer goods. An AI product image that can mislead people about the real product is a deep fake. Another is an advertising video in which simulated people demonstrate a product. A third is an advert featuring an AI-generated celebrity influencer. A real product shown against an AI background is not a deep fake, as long as the advert is unlikely to mislead people about the product. Colour correction or a new background for clearly aesthetic purposes is likely to have only a minor effect on whether people see an advert as authentic. AI voices for fictional characters, for example in audiobooks, games or animation, are not deep fakes, as long as nobody is misled about who is speaking.
Shop as deployer that publishes AI text to inform the public on matters of public interest (according to the guidelines, these include public health, environmental protection and consumer safety)The deployer must disclose that the text has been artificially generated or manipulated. No disclosure is needed if the text has been through human review or editorial control and someone holds editorial responsibility for it.According to the Commission's guidelines, the review should at least include a check of factual accuracy. A spell-check alone is not enough. AI-edited adverts and product descriptions fall outside this rule if they make no claims about matters such as health, consumer safety or sustainability.
Shop that only commissions an advertising agency and does not decide on the agency's use of AIAccording to the Commission's guidelines, the shop is not a deployer for that advert.The guidelines describe a company that commissions an advert but has no say in, and no control over, whether and how the agency uses AI.
Shop whose customer service staff use AI tools to answer customersAccording to the Commission's guidelines, this is not direct interaction under Article 50(1).The guidelines add that the mere possibility that a person could step in should not be used to get round the duty to give notice.

The two roles come from the AI Act, the examples from the Commission's guidelines. The guidelines add that providers may combine different notice techniques to suit the context, the audience and the resources available, especially to small and medium-sized enterprises.

An example

Take a hypothetical fashion shop that uses a customer service chatbot, a virtual try-on and AI product images. It also publishes an advertising video and product descriptions written with the help of AI.

The fashion shop case by case
SituationNotice needed?Source
The chatbot answers customer questionsYes; the provider of the chatbot must make sure the notice is given. The notice must come no later than the first interaction, in a clear and distinguishable manner. It must also meet the applicable accessibility requirements. According to the Commission's guidelines, the exception for obvious cases does not apply to helpdesk chatbots whose answers customers might mistake for a human's. The guidelines' example of a written notice is a chatbot whose first message says it is based on AI. Metadata or watermarks, which users cannot perceive during the chat, are not enough on their own. For AI on the phone, a spoken statement at the start of the call is one example of a suitable technique. If a shop developed the chatbot in-house and uses it under its own name, the guidelines treat the shop itself as the provider.Article 50(1) and (5) AI Act; Commission's guidelines
Customers use the virtual try-on for clothesNot under Article 50(1), according to the Commission's guidelines. They list virtual try-on among the systems that do not interact directly with people. That passage says nothing about marking or deep fakes.Article 50(1) AI Act; Commission's guidelines
An AI product image makes a dress look better made than it isYes; the shop, as deployer, must make the disclosure. According to the Commission's guidelines, such an image is a deep fake. The disclosure should be noticeable without technical aids, for example as a visible label. The shop cannot rely on the provider's machine-readable marking for this.Article 50(4) AI Act; Commission's guidelines
A real product photo against an AI background, with colours and image size adjustedUsually not. According to the Commission's guidelines, such an image is not a deep fake as long as it is unlikely to mislead people about the product. Colour correction, background changes for clearly aesthetic purposes and rescaling are likely to have only a minor effect on whether people see the image as authentic. This does not affect the provider's machine-readable marking, which Article 50(2) governs separately.Article 3(60) and Article 50(2) AI Act; Commission's guidelines
An advertising video shows a realistic AI influencer demonstrating a productYes; the shop, as deployer, must make a full disclosure. According to the Commission's guidelines, such a video is a deep fake and not an artistic work. The lighter disclosure for evidently creative or satirical works does not apply to content that is recognisably purely commercial.Article 50(4) AI Act; Commission's guidelines
Product descriptions written with the help of AINo marking by the shop. Under Article 50(2), machine-readable marking falls on the provider of the AI system, not on the shop that uses it. According to the Commission's guidelines, grammar and spelling correction, and translation, are exempt. Summarising or rewriting that changes style, structure and meaning needs marking by the provider. Product descriptions that make no claims about matters such as health, consumer safety or sustainability fall outside the rule for texts on matters of public interest.Article 50(2) and (4) AI Act; Commission's guidelines

These classifications follow the AI Act and the Commission's guidelines.

The rules in detail

The legal basis is the AI Act, Regulation (EU) 2024/1689 (original text). The transparency duties in Article 50 have applied since 2 August 2026. The consolidated version of the AI Act already includes the changes made by Regulation (EU) 2026/1744. The Commission published guidelines on Article 50 on 20 July 2026. The guidelines describe themselves as non-binding: only the Court of Justice of the European Union can interpret the AI Act authoritatively. The version we refer to is dated 5 November 2026; it was published in the Official Journal of the EU on 12 November 2026.

Key terms and provisions
ProvisionConcernsContent
Article 3(3)ProviderA provider is any person or body that develops an AI system, or has one developed, and places it on the market or puts it into service under its own name or trademark. This applies whether the system is supplied for payment or free of charge.
Article 3(4)DeployerA deployer is any person or body that uses an AI system under its authority, unless the use is part of a personal, non-professional activity. According to the Commission's guidelines, authority means taking responsibility for the decision to use the system and for how it is used, including its outputs. Technical control over how the system runs is not necessarily required.
Article 3(60)Deep fakeA deep fake is image, audio or video content generated or manipulated by AI that resembles existing persons, objects, places, entities or events and would falsely appear to a person to be authentic or truthful. According to the Commission's guidelines, realistic AI-generated avatars count as persons.
Article 50(1)ProvidersProviders must design and build AI systems intended to interact directly with people so that those people are told they are interacting with an AI system. No notice is needed where it is obvious to a reasonably well-informed, observant and circumspect person, given the circumstances and the context of use, that they are dealing with an AI system. According to the Commission's guidelines, this obviousness exception should be limited to cases where an average member of the intended and reasonably foreseeable audience has almost no doubt about what they are dealing with.
Article 50(2)ProvidersProviders of AI systems that generate synthetic audio, images, video or text must mark the outputs in a machine-readable format. The duty does not apply where the system only assists with standard editing. According to the Commission's guidelines, standard editing means small changes for readability, grammar, quality and format without creating new content.
Article 50(3)Deployers: emotion recognition and biometric categorisationDeployers of an emotion recognition system or a biometric categorisation system must inform the people exposed to it that the system is in operation. According to the Commission's guidelines, this covers any biometric categorisation system, for example one that sorts people by age or gender, unless the system is prohibited outright under Article 5(1)(g) of the AI Act.
Article 50(4), first subparagraphDeployers: deep fakesDeployers must disclose that a deep fake has been artificially generated or manipulated. Under the third sentence, a deep fake in an evidently artistic, creative, satirical, fictional or similar work needs only a suitable disclosure that such content is present, without spoiling the work.
Article 50(4), second subparagraphDeployers: text on matters of public interestDeployers must disclose that text published to inform the public on matters of public interest has been artificially generated or manipulated. No disclosure is needed if the text has been through human review or editorial control and someone holds editorial responsibility for it. Both conditions must be met.
Article 50(5)Providers and deployersThe information under paragraphs 1 to 4 must be given no later than the first interaction or exposure, in a clear and distinguishable manner. It must also meet the applicable accessibility requirements.

The table reflects the consolidated version of 27 July 2026, with explanations from the Commission's guidelines where marked.

A transition period covers generative AI systems placed on the market before 2 August 2026. Under Regulation (EU) 2026/1744, their providers have until 2 December 2026 to meet the marking duty in Article 50(2). The period does not apply to the notice under Article 50(1).

The voluntary Code of Practice on Transparency of AI-Generated Content was published on 10 June 2026, according to the Commission's FAQ on the code. On 8 July 2026, the Commission concluded in an opinion on the code that it adequately covers the duties on marking, disclosure and the form of notices. The same opinion says that following the code is not conclusive evidence of compliance. The Commission's FAQ on Article 50 adds that the code does not cover the notice under Article 50(1). The EU also offers icons for labelling AI-generated content. According to the Commission, using them is optional; the duties in Article 50 are not.

What to check

  • Where does AI in your shop talk directly to customers, for example a customer service chatbot, a voice assistant or an AI system that takes phone calls?
  • Did your shop develop its chatbot, or have it developed, and does it use the chatbot under its own name? That makes the shop the provider under the AI Act. The Commission's guidelines give a chatbot built in-house as an example.
  • Is the notice given no later than the first interaction, in a clear and distinguishable way that meets accessibility requirements? The Commission's guidelines give the example of a chatbot whose first message says it is based on AI.
  • Could AI images or videos mislead people about the real product, and if so, do they carry a visible label? According to the Commission's guidelines, the provider's invisible marking does not count as disclosure.
  • Has your shop modified another provider's generative AI system, and does it now use the new system under its own name? According to the Commission's guidelines, the shop then becomes the provider.
  • Do AI texts on matters of public interest, such as public health, go through human review, and does someone hold editorial responsibility for them? The Commission's guidelines expect at least a check of factual accuracy.

Sources

  1. Verordnung (EU) 2024/1689 (KI-Verordnung), Originalfassung vom 13. Juni 2024 (opens in a new tab)

    Amtsblatt der Europäischen Union (EUR-Lex) accessed

  2. Konsolidierte Fassung der Verordnung (EU) 2024/1689 (KI-Verordnung), Stand 27.07.2026 (opens in a new tab)

    Amt für Veröffentlichungen der EU (EUR-Lex/Cellar) accessed

  3. Künstliche Intelligenz: Leitlinien zu Transparenzpflichten veröffentlicht (20. Juli 2026) (opens in a new tab)

    Europäische Kommission, Vertretung in Deutschland accessed

  4. FAQ: Code of Practice on Transparency of AI-Generated Content (Stand 10.06.2026) (opens in a new tab)

    Europäische Kommission (Shaping Europe’s digital future) accessed

  5. EU Icons for labelling AI-generated content (Stand 13.10.2026) (opens in a new tab)

    Europäische Kommission (Shaping Europe’s digital future) accessed

History

  1. The transition period for machine-readable marking ends for generative AI systems placed on the market before 2 August 2026.
  2. The Commission's guidelines on Article 50 were published in the Official Journal of the EU (C/2026/6005), with the same wording as the version of 5 November 2026.
  3. The Commission's guidelines on Article 50, document C(2026) 7702 final, bear this date.
  4. The transparency duties in Article 50 of the AI Act began to apply.
  5. Regulation (EU) 2026/1744, which added the transition period for marking, was published in the Official Journal.
  6. The Commission published its guidelines on the transparency obligations under Article 50.
  7. The Commission concluded that the code of practice adequately covers the duties in Article 50(2), (4) and (5).
  8. The Code of Practice on Transparency of AI-Generated Content was published.
  9. The AI Act, Regulation (EU) 2024/1689, was adopted.

Guides on the topic

Glossary terms

Share on LinkedIn

Guidance, not legal advice: we name the document, the deadline and the primary source, and link to the other relevant sources. What applies in your case is for your legal adviser to judge.

Questions about implementing it in your shop?

We look at what the obligation means technically for your shop system and tell you plainly whether there is anything to build.

Search the website

Popular topics

All results