# EU AI Act Transparency Rules: What Website Chatbots and AI Content Need Now By [Eoin Villarin](https://amvprosox.com/about). Published 2026-10-29. Canonical: https://amvprosox.com/blog/eu-ai-act-article-50-transparency-chatbots-websites > Since 2 August 2026, EU websites must tell users when they talk to AI and label some AI content. Here is what Article 50 requires and how to comply. ## What do the EU AI Act transparency rules require for website chatbots? The EU AI Act transparency rules in Article 50 have applied since 2 August 2026. Chatbots and voice agents must tell people they are talking to AI unless it is obvious, deepfakes and unreviewed AI text on public-interest topics must be labelled, and generative AI output must carry machine-readable marking. Fines reach EUR 15 million or 3% of worldwide turnover. **Key takeaways** - Article 50 of the EU AI Act has applied since 2 August 2026 and was not postponed by the Digital Omnibus on AI. - The Digital Omnibus on AI, Regulation (EU) 2026/1744, entered into force on 27 July 2026 and moved Annex III high-risk rules to 2 December 2027. - Generative AI systems already on the market before 2 August 2026 have until 2 December 2026 to add machine-readable marking of their output. - Breaches of the AI Act transparency obligations can be fined up to EUR 15 million or 3% of worldwide annual turnover, whichever is higher, with lower caps for SMEs. - The Commission's Article 50 guidelines of 20 July 2026 say a chatbot disclosure placed only in terms and conditions is not enough. The EU AI Act transparency rules have applied since 2 August 2026, and they reach far more websites than the debate about high-risk AI suggests. If your site has an AI chatbot, a voice agent, AI-generated images of real people or places, or AI-written articles on topics like health or money, Article 50 of the AI Act now sets rules on telling people. The Digital Omnibus on AI, in force since July 2026, delayed the high-risk rules to December 2027, but it did not delay these. This guide is for EU online stores, service businesses, publishers and agencies, plus UK and US companies that serve EU customers. You get the rules in plain English, what the European Commission's new guidelines accept and reject, the fines, and a practical checklist for your chatbot, images and content. Dates, fines and examples come from the [AI Act text](https://eur-lex.europa.eu/eli/reg/2024/1689/oj), the Commission's [Article 50 guidelines](https://digital-strategy.ec.europa.eu/en/library/guidelines-transparency-obligations-providers-and-deployers-ai-systems) of 20 July 2026 and official EU institution pages, linked throughout. This post is general information, not legal advice. ## EU AI Act Transparency Rules at a Glance | Rule | Who it applies to | Date | | --- | --- | --- | | Tell people they are interacting with AI (Article 50(1)) | Providers of chatbots, voice agents, AI agents | 2 August 2026 | | Machine-readable marking of AI output (Article 50(2)) | Providers of generative AI | 2 August 2026, or 2 December 2026 for systems already on the market | | Disclose emotion recognition and biometric categorisation (Article 50(3)) | Deployers | 2 August 2026 | | Label deepfakes and unreviewed AI text on public-interest topics (Article 50(4)) | Deployers | 2 August 2026 | | High-risk stand-alone AI (Annex III) | Providers and deployers | Delayed to 2 December 2027 | | Maximum fine for transparency breaches | Companies | EUR 15 million or 3% of worldwide turnover, whichever is higher | ![Official EU icon reading AI GENERATED for labelling content under the EU AI Act transparency rules](https://amvprosox.com/api/img/1793276910707-kmwrt1) *The Commission's free "AI generated" icon for labelling content. Image: European Commission* ## What Changed With the Digital Omnibus on AI? The Digital Omnibus on AI amends the AI Act. The Council gave its [final approval on 29 June 2026](https://www.consilium.europa.eu/en/press/press-releases/2026/06/29/artificial-intelligence-council-gives-final-green-light-to-simplify-and-streamline-rules/), and it became Regulation (EU) 2026/1744, which [entered into force on 27 July 2026](https://artificialintelligenceact.eu/ai-act-explorer/digital-omnibus/). The changes that matter for most websites: - **High-risk rules moved.** Stand-alone high-risk systems listed in Annex III, such as AI used in hiring or credit scoring, now apply from 2 December 2027. High-risk AI embedded in regulated products moves to 2 August 2028. - **A short grace period for watermarking only.** Generative AI systems placed on the market before 2 August 2026 have until 2 December 2026 to add machine-readable marking under Article 50(2). The Commission's guidelines stress that chatbot disclosure for those same systems had to be in place from 2 August 2026. - **New bans.** AI systems that create non-consensual intimate images of identifiable people, or child sexual abuse material, are prohibited from 2 December 2026. - **Softer AI literacy duty.** Providers and deployers must now take measures to support AI literacy among their staff, rather than guarantee a specific level. ## Do You Have to Tell Visitors Your Chatbot Is AI? In most cases, yes. Article 50(1) requires providers of AI systems that interact directly with people to design them so people know they are talking to AI, unless that is obvious from the context. The Commission's guidelines list helpdesk and support chatbots on online platforms as an example where users may think they are talking to a person. ### Are you the provider or the deployer? This is the question most businesses get wrong. Under the guidelines, a company that develops a chatbot in-house, including on top of a model API, and puts it into service under its own name is a **provider**, so the disclosure duty is yours. If you take an existing generative AI system, modify it, for example with new training data, and run it under your brand, you also become a provider. If you install a third-party chat product that already carries its maker's disclosure, the maker is the provider and you are a deployer, but you should still check it actually shows the notice on your site. ### What a compliant disclosure looks like The guidelines are practical. They suggest a plain-language banner or a first-turn greeting that says the user is talking to an AI system, ideally with a persistent label near the input field. For voice agents, a spoken statement at the start of the call. A single prominent notice before the first interaction is likely enough in most cases, with reminders in riskier contexts or for vulnerable users. The guidelines also list what does not work on its own: a disclosure hidden in terms and conditions, an invisible watermark, a vague label such as "assistant", a human-like name and photo that could mislead, or a statement that only says the system uses LLMs. Even a site-wide line saying "Services on this website use AI" is called insufficient. The notice must also meet applicable accessibility rules, which ties directly to the [European Accessibility Act](https://amvprosox.com/blog/european-accessibility-act-website-compliance). ## Which AI Content Must Be Labelled? Article 50(4) puts two duties on deployers, which includes any business publishing AI content in the course of its work. ### Deepfakes AI-generated or manipulated images, audio or video that resemble existing people, objects, places or events and would falsely appear authentic must be disclosed. Artistic, satirical or fictional work gets a lighter duty that does not spoil the work. For a business site, this covers AI testimonial videos with realistic people, synthetic voices of real staff and realistic edited photos of real locations. ### AI-written text on matters of public interest Text generated or manipulated by AI and published to inform the public on matters of public interest must be labelled, unless a human reviewed it and a person or company holds editorial responsibility. The guidelines give useful examples. An AI-edited lifestyle article about diets and a disease is in scope. A company's ordinary ad copy or product descriptions are outside it, unless they make claims about health, consumer safety or sustainability. Human review means real review. The guidelines say fact-checking is the minimum, while spell-checks, an editorial policy that exists only on paper or a second AI checking the first do not count. The person or function holding editorial responsibility should be named with contact details somewhere easy to find, such as your legal or about page. ![Official EU icon reading AI MODIFIED for labelling partly AI-manipulated content](https://amvprosox.com/api/img/1793276911360-z64or1) *The "AI modified" variant is for content that was partly changed with AI. Image: European Commission* ### The free EU icons The Commission published a voluntary [Code of Practice on marking and labelling AI-generated content](https://digital-strategy.ec.europa.eu/en/news/commission-publishes-code-practice-marking-and-labelling-ai-generated-content) on 10 June 2026, together with a set of [EU icons](https://digital-strategy.ec.europa.eu/en/policies/eu-icons-labelling-ai-generated-content): a basic AI icon plus "AI generated" and "AI modified" versions in black and white. Anyone can use them without attribution. The Commission notes that using an icon does not by itself prove compliance, and that user testing found the icon works better with a text label. ## What Are the Fines Under Article 50? According to the Commission's guidelines, providers and deployers that breach the transparency obligations can be fined up to EUR 15 million or 3% of total worldwide annual turnover, whichever is higher. For SMEs and start-ups the cap is whichever is lower. National market surveillance authorities enforce the rules, and the AI Office supervises AI systems built on general-purpose models from the same provider. Anyone can file a complaint. The rules also reach beyond the EU. Providers placing AI systems on the EU market are covered wherever they are based, and non-EU deployers are covered when they direct AI output at EU audiences. ## What the EU AI Act Transparency Rules Mean for Your Website For EU online stores and service businesses, this is mostly front-end work plus a content process. Here is the checklist I use with clients. - **List every AI touchpoint.** Chat widgets, voice and phone agents, AI email replies, on-site image or text generators, AI avatars in videos and AI-assisted articles. - **Decide your role for each.** If you built it or run it under your brand, treat yourself as the provider. For third-party tools, ask the vendor in writing how it meets Article 50(1) and 50(2). - **Add a first-turn disclosure and a persistent label** to every chatbot, close to the input field, in the user's language. Drop human names and stock photos that suggest a person is typing. - **Make the notice accessible.** Use real text, sufficient contrast and an ARIA live region so screen readers announce it, in line with the [EAA fixes](https://amvprosox.com/blog/european-accessibility-act-website-compliance) your site already needs. - **Disclose handovers.** When a conversation moves between bot and human, say so. - **Label realistic AI media** of real people, places or products using the EU icons with a short text label, placed where people see it first. - **Set up an editorial process for AI-assisted content.** Fact-check before publishing, record who reviewed it, and name the person or role responsible on your site. Label anything on health, money or safety that skips this step. - **Mark AI output from tools you offer.** If your site lets customers generate images or text, add machine-readable marking such as C2PA metadata by 2 December 2026 at the latest. - **Train your team.** A short briefing on these rules covers the softened AI literacy duty and stops a marketing tool going live without a label. If your site is also being prepared for AI agents and AI search, the same audit pairs well with my guide to [making a website agent-ready](https://amvprosox.com/blog/agent-ready-website-structured-data-llms-txt-mcp). ## Frequently asked questions ### When did the EU AI Act transparency rules start? Article 50 of the AI Act has applied since 2 August 2026. The Digital Omnibus on AI did not postpone it. The only relief is a transition until 2 December 2026 for machine-readable marking by generative AI systems that were already on the market before 2 August 2026. Chatbot disclosure applied from 2 August. ### Does my website chatbot need an AI disclaimer under the EU AI Act? Yes, unless it is obvious the user is talking to AI. If you built the chatbot or run it under your own brand, you are likely the provider and responsible for the notice. The Commission recommends a first message or banner stating it is an AI system, often with a persistent label near the input box. ### Do AI-written blog posts need a label in the EU? Only some. Text published to inform the public on matters of public interest, such as health, finance or consumer safety, must be labelled unless a human reviewed it, including fact-checking, and someone holds editorial responsibility. Ordinary product descriptions and advertising are generally outside the rule unless they make health, safety or sustainability claims. ### What is the fine for breaking the AI Act transparency rules? Up to EUR 15 million or 3% of total worldwide annual turnover, whichever is higher, for companies. For SMEs and start-ups the cap is whichever amount is lower. National market surveillance authorities enforce the rules and must weigh the size and economic viability of smaller businesses when setting a fine. ### Does the EU AI Act apply to UK and US businesses? It can. Providers that place AI systems on the EU market are covered wherever they are based. Businesses outside the EU are also covered as deployers when they direct AI-generated content, such as deepfakes, at audiences in the EU. A UK or US store with an AI chatbot serving EU customers should follow Article 50. ## Small Labels, Real Obligations Most of the AI Act still feels distant for small businesses, but Article 50 is already in force and it touches the most common AI features on business websites. A clear chatbot notice, sensible labels and a real review step cover most of what regulators expect. If you want your chatbot, AI content and accessibility checked against these rules, or a compliant AI assistant built for your site, [send me a message](https://amvprosox.com/#contact-form) and I will review your setup and fix what needs fixing, working in European time zones. You can see examples of my work in my [case studies](https://amvprosox.com/case-studies). I will reply within 24 hours.