# EU AI Act Article 50: transparency obligations, paragraph by paragraph

> With the official wording: who Article 50 applies to (providers or deployers), what online stores must disclose since 2 August 2026, and the penalties.

URL: https://easyaiactlabels.com/en/guides/eu-ai-act-article-50/

[Guides](/en/guides/) /Obligation

# EU AI Act Article 50: transparency obligations, paragraph by paragraph

Last updated: 4 October 2026 Responsible: Martin Jäger Created with AI 9 min read

Short answer

Article 50 of the EU AI Act sets out transparency obligations for certain AI systems in seven paragraphs and has applied since 2 August 2026. For an online store as a deployer, two parts matter: Article 50(4) (disclose AI images, video and audio that look real; text only where it informs the public on matters of public interest) and Article 50(5) (clear, distinguishable, at the latest at first exposure, accessible). Article 50(1) on chatbots and Article 50(2) on machine-readable marking are addressed to providers, Article 50(6) leaves other transparency rules such as unfair competition law untouched, and Article 50(7) covers codes of practice. The Digital Omnibus (Regulation (EU) 2026/1744) only replaced paragraph 7.

Article 50 is the provision of the EU AI Act (Regulation (EU) 2024/1689) that creates the AI labelling obligation. It has applied since 2 August 2026 (Article 113 AI Act). This guide walks through it paragraph by paragraph and says what each part means for an online store. The practical question of which images need a label is covered in [EU AI Act Article 50: do I have to label AI images in my online store?](/en/guides/ai-images-labelling-obligation/)

## What does Article 50 of the AI Act regulate?

Article 50 is headed “Transparency obligations for providers and deployers of certain AI systems”. **Providers** develop an AI system, or have it developed, and place it on the market or put it into service under their own name (Article 3(3)). **Deployers** use an AI system under their own authority in a professional context (Article 3(4)). A store that creates images or text with ready-made tools is a deployer.

Paragraph

Addressed to

Core

For your store

50(1)

providers

AI that interacts with people must disclose that it is AI

in practice yes, once you use a chatbot

50(2)

providers

mark outputs in a machine-readable format

no

50(3)

deployers

inform about emotion recognition and biometric categorisation

usually no

50(4), first subparagraph

deployers

disclose deep fakes (image, audio, video)

**yes**

50(4), second subparagraph

deployers

disclose AI text on matters of public interest

rarely

50(5)

all

clear, distinguishable, at first exposure, accessible

**yes**, as the rule on form

50(6)

nobody

other transparency obligations remain unaffected

yes, the bridge to unfair competition law

50(7)

Commission

codes of practice

indirectly, through the Code of Practice

## Which version applies?

Article 50 still reads as adopted in 2024, with one exception: the Digital Omnibus (Regulation (EU) 2026/1744 of 8 July 2026, in force since 27 July 2026) replaced paragraph 7 (Article 1, point 20). Paragraphs 1 to 6 and the definition of deep fake in Article 3(60) are unchanged. The Omnibus also added Article 111(4): providers whose systems were placed on the market before 2 August 2026 have until 2 December 2026 to comply with Article 50(2) (Article 1, point 39(b)). On 4 October 2026, no consolidated version was available from the data service of the EU Publications Office, so anyone quoting the Act reads the 2024 text and the Omnibus side by side.

## Article 50(1): chatbots and other interactive AI

> “Providers shall ensure that AI systems intended to interact directly with natural persons are designed and developed in such a way that the natural persons concerned are informed that they are interacting with an AI system, unless this is obvious from the point of view of a natural person who is reasonably well-informed, observant and circumspect, taking into account the circumstances and the context of use.”

Article 50(1) is addressed to providers (Guidelines, paragraph 28). In practice it reaches your store as soon as an AI chat widget answers customers there: the notice has to stay visible in use. The guidelines recommend a clear notice at the start of the conversation, such as a first-turn greeting, and a label close to the input field (paragraph 37). Not sufficient on their own: a notice only in the terms and conditions, machine-readable markings, vague names such as “assistant”, or general statements like “Services on this website use AI” (paragraph 38). More in [AI texts and AI chatbots in your online store](/en/guides/ai-texts-chatbots-labelling/).

## Article 50(2): machine-readable marking by the provider

> “Providers of AI systems, including general-purpose AI systems, generating synthetic audio, image, video or text content, shall ensure that the outputs of the AI system are marked in a machine-readable format and detectable as artificially generated or manipulated.”

This obligation falls on the makers of the generators, not on you. Their technical solutions must be effective, interoperable, robust and reliable “as far as this is technically feasible”. Systems that only perform an assistive function for standard editing, or that do not substantially alter the input data, are exempt; the guidelines list examples such as minor colour adjustments, minor cropping, noise reduction and format conversion (paragraphs 90 to 92).

According to the guidelines, marking always comes with a way to detect it (paragraphs 69 and 70). For signatories, the Code of Practice of 10 June 2026 sets out two layers: digitally signed metadata (sub-measure 1.1.1) and an imperceptible watermark (sub-measure 1.1.2), plus a detection tool free of charge (sub-measure 2.1.1). What this means for you is covered in [AI watermarks](/en/guides/ai-watermark/); which generators already mark their output is in [Which AI image generators write provenance evidence?](/en/guides/ai-image-generators-provenance/)

## Article 50(3): emotion recognition and biometric categorisation

Deployers of an emotion recognition system or a biometric categorisation system must inform the people exposed to it. In everyday store work with AI images and text, this usually does not come up.

## Article 50(4), first subparagraph: disclose deep fakes

> “Deployers of an AI system that generates or manipulates image, audio or video content constituting a deep fake, shall disclose that the content has been artificially generated or manipulated.”

This is the rule that applies to your store. Article 3(60) defines a deep fake as “AI-generated or manipulated image, audio or video content that resembles existing persons, objects, places, entities or events and would falsely appear to a person to be authentic or truthful”. The guidelines test this in four steps (paragraph 113): resemblance to the subject; the subject exists or could plausibly exist; it is a person, object, place, entity or event, expressly including consumer goods; and the content would falsely appear authentic, judged objectively by context and audience.

Photorealistic AI product images, AI models and AI lifestyle scenes are therefore regularly covered. AI edits of insignificant details are usually not, as long as the ad does not mislead about the product: for product advertising, the guidelines give colour correction, background extensions, replacing backgrounds for clearly aesthetic purposes and rescaling as examples likely to have only a minor impact, depending on the context (paragraph 116).

Where the content forms part of an evidently artistic, creative, satirical or fictional work, the obligation is limited to a disclosure that does not hamper the display or enjoyment of the work (third sentence of Article 50(4)). This rarely helps stores: content of a solely commercial character that is recognisable as such does not qualify, and advertising only in certain specific situations (paragraphs 120 to 122). Which images are affected in detail: [Which images in your store need an AI label?](/en/guides/which-images-need-a-label/)

## Article 50(4), second subparagraph: AI text

> “Deployers of an AI system that generates or manipulates text which is published with the purpose of informing the public on matters of public interest shall disclose that the text has been artificially generated or manipulated.”

The obligation does not apply where the AI-generated content “has undergone a process of human review or editorial control” **and** a natural or legal person holds editorial responsibility for the publication. Both are needed; a spell check or an approval click without looking at the content is not enough (paragraphs 134 and 135). The guidelines list product descriptions and advertising copy as an example outside the obligation, except for claims related to, for example, health, consumer safety or sustainability.

## Article 50(5): form and timing

> “The information referred to in paragraphs 1 to 4 shall be provided to the natural persons concerned in a clear and distinguishable manner at the latest at the time of the first interaction or exposure. The information shall conform to the applicable accessibility requirements.”

Clear means noticeable, easy to understand and accessible, including for persons with disabilities; distinguishable means easy to identify as separate from other information and its surroundings (Guidelines, paragraph 142). A notice in a manual, under layers of menus or in terms of use is not enough. The obligation applies to each output with respect to every person who sees it for the first time (paragraph 143). Labels must be perceivable without technical tools, visible or audible, which is why the machine-readable marking under Article 50(2) does not suffice (paragraph 117). Article 50 sets no accessibility rules of its own; it refers to those that apply (paragraph 144, more in [Is your AI label accessible?](/en/guides/accessible-ai-label/)). What correct labels look like is shown in [AI generated label examples](/en/guides/ai-generated-label-examples/).

## Article 50(6): other obligations remain

> “Paragraphs 1 to 4 shall not affect the requirements and obligations set out in Chapter III, and shall be without prejudice to other transparency obligations laid down in Union or national law for deployers of AI systems.”

For stores, this is the bridge to unfair competition law. An AI image that shows a product differently from what is delivered is misleading, with or without a label; the guidelines make clear that disclosure does not make misleading advertising lawful (paragraph 129).

## Article 50(7): codes of practice

Since the Omnibus, the Commission encourages codes of practice on the detection, marking and labelling of artificially generated or manipulated content and assesses whether they are adequate for Article 50(2) and (4); if it deems a code inadequate, it may adopt common rules. The code in question is the Code of Practice on Transparency of AI-generated Content of 10 June 2026, with one section for providers and one for deployers. It is voluntary and does not create a presumption of conformity (recital 41 of Regulation (EU) 2026/1744). What Section 2 means for you: [The EU Code of Practice on AI labelling](/en/guides/code-of-practice-ai-labelling/).

## What are the penalties for breaching Article 50?

Article 99(4)(g) of the AI Act names the transparency obligations of Article 50 expressly: fines of up to EUR 15 million or 3% of worldwide annual turnover, and for SMEs whichever of the two is lower (Article 99(6)). In Germany, the Federal Network Agency (Bundesnetzagentur) is the competent authority (§ 2 KI-MIG). In practice, a warning letter under unfair competition law is the more likely risk, see [Who enforces the EU AI image labelling rule in your store?](/en/guides/enforcement-and-violations/) and [Warning letter over an unlabelled AI image](/en/guides/warning-letter-ai-image-labelling/)

## How binding are the guidelines and the code?

The Commission guidelines of 20 July 2026 (Annex to C(2026) 5054 final) are non-binding; only the Court of Justice of the EU can give an authoritative interpretation (paragraph 5). They remain the most important yardstick, because market surveillance authorities and courts are likely to look to them. For signatories of the code, supervision focuses on whether they follow the code (paragraph 147); non-signatories are expected to explain their own measures (paragraph 148). As of 30 September 2026, no court decision on labelling AI images under Article 50 is known in Germany, Austria or Switzerland.

Labelling remains your decision and your responsibility as a deployer.

## Frequently asked questions

### Which paragraph of Article 50 applies to my online store?

Mainly Article 50(4), first subparagraph: as a deployer, you disclose that image, audio or video content that looks real has been artificially generated or manipulated. Article 50(5) adds the rules on form and timing. Article 50(4), second subparagraph (text), only applies to text on matters of public interest; Article 50(1) (chatbots) and Article 50(2) (marking) are addressed to providers.

### What is the difference between Article 50(2) and Article 50(4)?

Article 50(2) requires providers of AI generators to mark outputs in a machine-readable format, for example with signed metadata or invisible watermarks. Article 50(4) requires deployers such as your store to disclose realistic AI content visibly to people. You cannot rely on the Article 50(2) marking to meet Article 50(4) (Guidelines, paragraph 117).

### Did the Digital Omnibus postpone Article 50?

No. Regulation (EU) 2026/1744 only replaced Article 50(7) (Article 1, point 20). It added a transition period until 2 December 2026 for providers of older generators, and only for Article 50(2) (new Article 111(4)). The store's obligation under Article 50(4) has applied since 2 August 2026.

### Are the Commission guidelines on Article 50 binding?

No. The guidelines of 20 July 2026 are expressly non-binding; only the Court of Justice of the EU can give an authoritative interpretation (paragraph 5). They come from the body that coordinates enforcement, so they are still the most important interpretation aid.

### Do I have to sign the Code of Practice?

No. The code is voluntary and, according to recital 41 of Regulation (EU) 2026/1744, does not create a presumption of conformity. The Commission expects non-signatories to explain on request how their measures comply with the obligation (Guidelines, paragraph 148).

## Sources

1.  Regulation (EU) 2024/1689 (AI Act), CELEX 32024R1689, Articles 3(4), 3(60), 50, 99, 113, recitals 132 to 136. [publications.europa.eu](http://publications.europa.eu/resource/celex/32024R1689) (accessed 4 October 2026)
2.  Regulation (EU) 2026/1744 (Digital Omnibus on AI), CELEX 32026R1744, Article 1 points 20, 38, 39(b), 40; recitals 38 and 41. [publications.europa.eu](http://publications.europa.eu/resource/celex/32026R1744) (accessed 4 October 2026)
3.  European Commission, Guidelines on the implementation of the transparency obligations under Article 50, Annex to C(2026) 5054 final of 20 July 2026. [ec.europa.eu](https://ec.europa.eu/newsroom/dae/redirection/document/131215) (accessed 4 October 2026)
4.  Code of Practice on Transparency of AI-generated Content of 10 June 2026. [ec.europa.eu](https://ec.europa.eu/newsroom/dae/redirection/document/129555) (accessed 4 October 2026)

This article is general information, not legal advice. Whether an image needs a label is your decision; for your specific case, please ask a lawyer.

Read next

[03 · Obligation What changes on 2 December 2026 for AI images in your store?](/en/guides/ai-images-2-december-2026/) [25 · Obligation Labelling AI videos in your online store and in ads: product videos, avatars, AI voices](/en/guides/ai-videos-labelling/)

I find the AI evidence. You decide with one click.

[View on the Shopify App Store](https://apps.shopify.com/easy-ai-act-image-labels?utm_source=easyaiactlabels&utm_medium=website&utm_content=artikel)