Which AI disclosure rules apply to me?

Points about EU Article 50 that change the answer

The finder lists five parts of Article 50 of the EU AI Act. Five points decide how it reads for a business like yours. Most come from the Commission's Guidelines on Article 50 (C(2026) 5054 final, published July 20, 2026), which are non-binding (paragraph 5): only the Court of Justice of the EU gives an authoritative interpretation of the Act. We verified each point on October 21, 2026.

• Provider or deployer. The chatbot duty in Article 50(1) is addressed to the provider of the system. A company that supplies a chatbot under its own name or trademark is the provider, including when its customers use it as supplied, without changes (paragraphs 10 and 11). On that reading a business that embeds the vendor's chatbot is a deployer with no Article 50(1) duty of its own, while one that builds, commissions or changes a chatbot and runs it under its own name or trademark is the provider. The label on a deep fake is a deployer duty (Article 50(4)), and a deployer cannot rely on the machine-readable marking the provider embeds, because people cannot perceive it (paragraph 117).
• Commercial imagery. The Guidelines list an AI-generated product image in an advertisement or on packaging as a deep fake when it can mislead about the real product's appearance, characteristics or use, for example by making it look more appealing or of better quality than in real life. They list a real product shown against an AI-generated background as not a deep fake, as long as the ad is not likely to mislead, and they expect AI color correction, background extensions, background replacement for clearly aesthetic purposes, arrangements of existing products and rescaling to have only a minor impact (paragraph 116). The Commission's page on the EU icons lists "authentic photographs of an empty apartment is furnished using AI" among the use cases for its Partially AI-Modified icon; the Guidelines themselves do not settle whether virtual staging is a deep fake.
• Dates. Article 50 applies from August 2, 2026, whatever the date a system was placed on the market (paragraph 153). Regulation (EU) 2026/1744, published in the Official Journal on July 24, 2026, adds a transition for one duty only: providers of generative AI systems placed on the market before August 2, 2026 have until December 2, 2026 to meet the machine-readable marking duty in Article 50(2) (Article 111(4)). It is for providers, not for the businesses that use their tools, and the chatbot disclosure and the deep fake label were due on August 2, 2026. Content generated before that date does not need to be labeled retroactively, but text generated earlier and published on or after it does need a label where the text duty applies (paragraph 154).
• Fines. Up to EUR 15,000,000 or, for a company, 3 percent of total worldwide annual turnover, whichever is higher. For SMEs and small mid-caps the cap is whichever is lower (Article 99(4), 99(6) and 99(6a)). We found no enforcement action under Article 50 as of October 21, 2026, which is not proof there is none.
• Accessibility. Article 50(5) requires the information to conform to the applicable accessibility requirements. The Commission says Article 50 adds no distinct or additional accessibility requirement (paragraph 144), so the question is which accessibility rules already apply to you. The European Accessibility Act exempts microenterprises from its service requirements: fewer than 10 persons and an annual turnover or balance sheet total not over EUR 2 million (Directive 2019/882, Article 3(23)). Our guide to accessible AI labels covers what a label can do for screen reader users either way.

How to read a result

Each result says why it was listed, who the rule binds, what triggers it, what it asks for, with the wording quoted where a law prescribes some, its key dates and its penalty. Items we could not read in the primary source are listed as not verified, and every result links to its source and to the GotAlt page that explains it. The finder lists rules as ones that "may apply", never as ones that apply. When nothing matches, the finder says that no specific rule matched and points to our overview of AI disclosure laws.

The rules this finder checks

The table lists every rule the finder can show, built from the same data the results use, so the two cannot differ. Laws, platform rules and voluntary standards are marked as such. Status is as of October 21, 2026, the date we verified each entry. Our pages on EU AI Act Article 50, US AI disclosure laws, platform AI disclosure rules and China and South Korea go deeper.

What the finder does not cover

Other countries and other US states have their own rules, and the questions leave out many situations: companion chatbots, election and political ads, the EU AI Act's other duties such as those for high-risk systems, and the consumer protection, personality rights, data protection and copyright law that a label does not displace. Where we could not read a source, the result says so. For anything the finder leaves out, start with our overview of AI disclosure laws.

After you know which rules may apply

The accessible AI label generator drafts label wording for several of these rules, the AI label and chatbot disclosure checker looks at a live page, and the AI image metadata checker and AI image metadata writer read and add the markers a file declares. They read what a file or page declares; none of them decides whether an image is AI-generated, and a missing marker proves nothing. Our guides cover accessible AI labels, alt text for AI-generated images, virtual staging disclosure, AI chatbot disclosure, synthetic performer ads and AI image metadata.

No tool makes a site or a label meet a law, and neither does an icon: the Commission says use of its EU icons does not establish legal compliance by itself. What we check lists what our scans cover, and overlays and the FTC explains why a widget does not change the answer.