# Can I use the EU AI Act's labelling rules as a reason to require AI disclosure inside my company? Source: https://global100.org/forum/q/can-i-use-the-eu-ai-acts-labelling-rules-as-a-reason-to-require-ai/ Site: Global 100 Forum, category Policy, law and disclosure Published: 2026-10-18. Updated: 2026-10-21. Replies: 8. ## Question by Ingrid V., 2026-10-18 I run trust and safety for a marketplace with sellers and buyers in the EU and the US. I am drafting an internal policy that would require anyone using generative AI for seller-facing content to disclose it: support macros, listing templates, help centre articles, marketing emails, policy text. Internal disclosure in every case, an external label where the content could be mistaken for a human statement, and a human owner named on each piece. I wanted to anchor the policy in Article 50 of the EU AI Act, since the transparency obligations started applying in August. Our legal team pushed back. Their view is that Article 50 does not require most of what I am proposing, so citing it as the reason would be misleading, and the policy should stand on our own values instead. I think they are probably right on the letter, but "our values" is a weak foundation when a product manager wants an exception. So, for people who have read the text rather than the summaries: what does Article 50 actually require of a company like ours, as a deployer rather than a model provider? Is anything else in the Act a legitimate hook for an internal disclosure rule? And is "the law requires this" an honest way to sell the policy internally, or a stretch that will be found out? ## Reply 1 by Kwame A., 2026-10-18 Your legal team is right on the letter, and an honest framing is available without stretching anything. [Article 50](https://artificialintelligenceact.eu/article/50/) puts its duties on different actors. Paragraph 1: providers of systems that interact directly with people must make sure the person knows it is an AI, unless obvious to a reasonably well-informed, observant and circumspect person. Paragraph 2: providers of systems generating synthetic audio, image, video or text must mark outputs in a machine-readable, detectable way, as far as technically feasible. Paragraph 4 is the deployer paragraph: disclose deep fakes you deploy, and disclose AI text "published with the purpose of informing the public on matters of public interest", unless it has had human review or editorial control and a person holds editorial responsibility. Listing templates and marketing emails are not public-interest text, so Article 50 does not require disclosure on them. It gives you a floor and a vocabulary. "Article 50 sets the minimum; our policy goes above it, using its categories" is accurate and still carries weight. The honest hook for staff-facing rules is elsewhere; more below. ## Reply 2 by Sam Whitlock, 2026-10-19 Practical angle: the Act requires AI output to be watermarked, so on a marketplace your real compliance job is detection. Run everything sellers submit through a detector, flag what the watermark check catches, and the disclosure policy more or less enforces itself. Build the pipeline before the policy. ## Reply 3 by Tobias K., 2026-10-19 On drafting, Ingrid, since this is where the exceptions will be fought. Our institution went through the same exercise this year and the structure that survived committee had three parts, kept visibly separate in the document. Part one: what the law requires of us, with citations and nothing added. Part two: what we require of ourselves, stated as our decision, with the reasons. Part three: the procedure, meaning who discloses what to whom, in what form, and what happens on non-compliance. The product manager seeking an exception then has to argue against part two on its merits, in front of the people who decided it, rather than against a legal claim that turns out to be inflated. In my experience the second argument is lost more often than the first, because the person has to say out loud that the company's stated reasons do not apply to them. One further suggestion: define your disclosure classes in part three before anyone has a live case, so that the first exception request is handled against a written standard rather than negotiated. ## Reply 4 by Hana Sato, 2026-10-19 @Sam Whitlock two corrections, offered gently. The marking duty in paragraph 2 sits on the provider of the generating system, not on a deployer or a platform hosting the results, and it is qualified by "as far as technically feasible". Nothing in Article 50 asks Ingrid's company to detect anything. Second, a detector is not a watermark check. Watermark verification needs the provider's key or scheme and a mark that survived copying and editing; a general detector is a classifier guessing from style. At marketplace scale a classifier-based enforcement regime would produce a steady stream of false flags on honest sellers, many of them the non-native writers the [bias thread](/forum/q/are-ai-detectors-biased-against-non-native-english-writers/) is about. The [watermarking thread](/forum/q/does-watermarking-ai-generated-text-actually-work/) goes through why text marks in particular are fragile. Policy first, and no detector as the enforcement arm. ## Reply 5 by Lena B., 2026-10-21 Seller side of this, from the freelance world. When a platform says "disclose AI use", two things happen. The honest sellers disclose and get fewer sales, because buyers read the label as a warning. The others do not disclose and get away with it, because nobody can tell. You have now punished honesty. If you want disclosure to work, make the label neutral and make the consequence for non-disclosure real and about something you can actually prove: misrepresentation found in a dispute, not a detector score. Otherwise the policy selects for liars. I have watched it happen on three platforms. ## Reply 6 by Marco Ferri, 2026-10-21 The exemption in paragraph 4 is the part I would borrow wholesale. For public-interest text the Act lifts the disclosure duty when a human has reviewed the content and a natural or legal person holds editorial responsibility. That is how a newsroom already works: a named editor signs off, and the byline is a promise about who stands behind the words. Your "human owner named on each piece" is the same idea, and it is the mechanism that scales. A label tells the reader a model was involved. An owner tells them who answers for it. Ingrid, I would make the owner mandatory in every class and the label conditional, rather than the other way round. The owner is what you will need when something published under your name turns out to be wrong. ## Reply 7 by Ingrid V., 2026-10-21 Kwame, that is the framing I needed: Article 50 as the floor and the vocabulary, our policy as the decision above it. You said the honest hook for staff-facing rules is elsewhere. Where? And does the chatbot duty in paragraph 1 catch our support assistant, which answers sellers in the help centre? ## Reply 8 by Kwame A., 2026-10-21 @Ingrid V. two answers. The hook is [Article 4](https://artificialintelligenceact.eu/article/4/), AI literacy, which has applied since 2 February 2025. Providers and deployers must "take measures to support the development of AI literacy of their staff and other persons dealing with the operation and use of AI systems on their behalf", taking account of their training and the context of use. It does not require any particular level of literacy, but a written disclosure policy with training is a very natural "measure", and that is a defensible sentence to put in part one of Tobias's structure. On the support assistant: paragraph 1 is a provider duty, so formally it sits with whoever built the system, but you are the one facing the seller. If a reasonably observant person would not realise they are talking to an AI, a clear notice at the start of the conversation solves it for everyone. Timing is settled: Article 50 applies from 2 August 2026 under Article 113, so this is current law, not a future one. The [disclosure thread](/forum/q/do-you-have-to-disclose-ai-generated-content-and-who-requires-labels/) has the platform rules alongside it. --- Cite as: Global 100 Forum, "Can I use the EU AI Act's labelling rules as a reason to require AI disclosure inside my company?", https://global100.org/forum/q/can-i-use-the-eu-ai-acts-labelling-rules-as-a-reason-to-require-ai/, accessed 2026-11-10.