AI Act Art. 50: What Your UI Must Show Since 02.08.2026 (Fines up to 15M)
Art. 50 has applied since 02.08.2026: chatbot disclosure, machine-readable marking, deepfake label, emotion recognition notice. Fines up to 15M EUR or 3%.
Since 02.08.2026 the transparency obligations of Art. 50 apply to certain AI systems and uses. The European Commission announced on 31 July 2026 that from 2 August 2026 its AI Office and the national authorities begin enforcing the AI Act. Chatbots and other interactive AI systems must tell users they are dealing with AI, deepfakes must be labelled, and AI-generated content within the scope of Art. 50(2) must carry machine-readable marks. Breaches of Art. 50 can be fined up to 15M EUR or 3% of worldwide annual turnover, whichever is higher. Not 35M, that is Art. 5. Art. 50 contains four obligations addressing providers and deployers. The fine ceiling is a maximum, not a fixed amount.
As of October 2026
What Art. 50 actually demands
Four paragraphs, four obligations. Which of them apply to you depends on your role and your use case. Short form first, detail after.
| Paragraph | Addressee | Obligation | Typical trigger |
|---|---|---|---|
| Art. 50(1) | Provider | AI system that interacts directly with humans must be recognisable as AI | Chatbot, voice assistant, AI companion, copilot |
| Art. 50(2) | Provider | Synthetic audio, image, video, text output must be machine-readably marked as AI-generated | Image generator, voice cloning, GPT output |
| Art. 50(3) | Deployer | Emotion recognition or biometric categorisation requires informing affected persons | Call centre sentiment analysis from voice |
| Art. 50(4) | Deployer | Deepfake image, audio, video must be labelled as artificially generated; so must published AI text on matters of public interest without human review or editorial control | Avatar or voice of a real person in a commercial, deceptively realistic AI images |
Per Art. 50(5) the information must be delivered "in a clear and distinguishable manner at the latest at the time of the first interaction or exposure". That is the legal deadline. Our advice beyond it: show the notice earlier, not only in the footer, the ToS or the cookie banner scroll.
These obligations have applied since 02.08.2026. The only grace period covers the marking obligation in Art. 50(2), and only for systems placed on the market before 02.08.2026: they have to meet it from 02.12.2026 (European Commission Art. 50 FAQ, updated 24 July 2026).
Provider vs. Deployer: who labels what
The most common mix-up in the DACH Mittelstand. The obligations of paragraphs 1 and 2 hit the provider, meaning whoever 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 (Art. 3(3) AI Act). The obligations of paragraphs 3 and 4 hit the deployer, meaning whoever uses the AI system professionally.
Concretely: if you use a ready-made AI system from a vendor under your own responsibility, for example a bought chatbot in your customer support UI, you are usually the deployer; the vendor of the system is the provider. The obligation in Art. 50(1) (make AI recognisable as such) is addressed to the provider. As a deployer you carry the duties in Art. 50(3) and (4): emotion recognition, deepfakes and AI text on matters of public interest. It is different if you build your own system on top of an AI model's API and offer it under your own name or trademark: that can make you the provider of that system, the provider of the model is not automatically the provider of your system. (Note: the value-chain rules in Art. 25 and the deployer duties in Art. 26 only kick in once the system counts as high-risk (Annex III or Annex I, deadlines below), typically not the case for a vanilla customer support bot.)
If you build an AI system yourself or have one built and place it on the market under your own name or trademark, for example a fine-tuned open-source LLM as a product feature or generative AI as white-label inside your product: you are the provider (Art. 3(3) AI Act). You then carry the provider duties in Art. 50(1) and (2), and if you also use the system yourself, the deployer duties in Art. 50(3) and (4) on top.
Practical rule of thumb: whoever offers an AI system under their own name or trademark is the provider. Whoever uses someone else's system under their own authority is the deployer. Check the role case by case.
Art. 50(1): Chatbot outing
The original text is sober: providers must ensure natural persons are "informed that they are interacting with an AI system". Exception: when this is obvious from the point of view of a reasonably well-informed, observant and circumspect natural person, taking into account the circumstances and the context of use.
The Commission reads this exception restrictively, because it deprives people of transparency (Art. 50 FAQ). A chatbot with a robot avatar and the name "Bot Assistant Lisa" is not automatically obvious, because many customer support UIs use staff avatars and first names. Voice assistants that clone human voices are hardly obvious. The notice must be given clearly and distinguishably from the start of the first interaction. AI that works only in the background or through machine-to-machine communication, without direct contact with people, is not covered (Art. 50 FAQ).
What is advisable in practice (beyond the statutory text):
- Persistent "AI assistant" badge in the conversation UI, not just at the start.
- Plain language: "You are chatting with an AI" or "AI Assistant", not "virtual employee" or "smart agent".
- For voice: explicit announcement at the start and before every sub-flow that could imply a human (e.g. complaint escalation).
- Audit trail: logging that the notice was served, with timestamp.
Further exception: AI systems authorised by law for law enforcement (Art. 50(1) sentence 2, details in the section on exceptions).
Art. 50(2): Synthetic content machine readability
This gets technical. Providers of AI systems that generate synthetic audio, image, video or text output must ensure that output is "marked in a machine-readable format and detectable as artificially generated or manipulated". As far as technically feasible, the techniques must be effective, interoperable, robust and reliable (Art. 50 FAQ).
The statutory text names no specific format. What matters in practice is the Code of Practice on Transparency of AI-generated Content: the final version was published on 10 June 2026. Section 1 addresses providers (marking and detection), Section 2 addresses deployers (labelling of deepfakes and AI text). The Commission and the AI Board have confirmed the code as an adequate voluntary tool to demonstrate compliance with Art. 50. According to the Commission, about 190 companies and organisations had signed by the end of July 2026. Those who do not sign must demonstrate compliance through other, equally adequate means and may face more requests for information from authorities.
C2PA (Coalition for Content Provenance and Authenticity) is a widely used open industry standard that embeds cryptographically signed provenance information into files. It is one possible technique, not a requirement of the law. How the obligations can be met technically for each content type is explained by the Code of Practice and the Commission's guidelines on Art. 50.
Exception: the system only performs an assistive function for standard editing. Example: auto-correct in Word is not Art. 50(2), an AI-generated image from Midjourney is. The Commission names further exclusions, for instance short strings of characters and machine-to-machine outputs without exposure to humans.
What you concretely need (as provider):
- Machine-readable marking of outputs within the scope of Art. 50(2), for example via a C2PA manifest or watermark, following the Code of Practice or an equivalent approach.
- A way to detect the marking, because the obligation covers detectability of the output.
- For systems placed on the market before 02.08.2026: implementation by 02.12.2026.
What you need (as deployer): nothing directly from paragraph 2, but paragraph 4 often kicks in once you forward the output to end customers.
EU AI Act
AI literacy under Article 4: build it and document it
Art. 50(3): Emotion recognition disclosure
Whoever deploys an emotion recognition system or a biometric categorisation system must inform the affected persons of the operation of the system, and must process the data per GDPR.
Sounds narrow, is not. What is emotion recognition? Any system that infers emotions or intentions from biometric data. Voice sentiment analysis in a call centre: yes. Sentiment from facial expression in a sales call: yes. Caution: Art. 5 in principle prohibits emotion recognition in the workplace and in education institutions, with narrowly defined exceptions. For HR tools a notice under Art. 50 is therefore not enough.
Biometric categorisation: assigning natural persons to categories based on biometric data. If a system estimates age, gender or mood from a photo for marketing targeting: yes. Here Art. 50 overlaps with the prohibition list in Art. 5: biometric categorisation by sensitive characteristics is in principle prohibited under Art. 5(1)(g) (with narrow exceptions). If your system falls under that, Art. 5 applies, with fines up to 35M EUR or 7%.
What you concretely need:
- Our recommendation: a notice before the interaction that discloses the sentiment tool ("This conversation is automatically analysed for tonality"). Legally, the information must be given at the latest at the time of the first interaction or exposure.
- GDPR duties in parallel: legal basis under Art. 6 GDPR, notice per Art. 13 GDPR, deletion concept.
- An opt-out path is advisable. More in our GDPR + Agentic AI Production guide.
Art. 50(4): Deepfake labelling
Deployers who generate or manipulate deepfake image, audio or video content must disclose that the content was artificially generated. This applies regardless of whether you are provider or just using a tool.
This is the paragraph that marketing departments can trip over. Under Art. 3(60) AI Act a deepfake is 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. Whether that applies also depends on context and audience expectations (Art. 50 FAQ). Concrete cases:
- AI-generated avatar or cloned voice of a real person (for example the CEO) in a corporate film or explainer video that looks or sounds real: deepfake label needed.
- Realistic-looking, entirely invented people can be covered as well, because the Commission also accepts things that "could plausibly exist". When in doubt, label.
- AI images in the blog header: whether this is a deepfake depends on whether the image is taken as real. The provider's marking duty under paragraph 2 is unaffected.
For labelling, the EU provides a set of icons that deployers can use.
The crucial exception: artistic, creative, satirical, fictional. But even then the existence of the generated content must be disclosed in an appropriate manner that does not hamper the display or enjoyment of the work. A discreet note that does not disturb the work can suffice. Complete silence does not.
Often forgotten second subparagraph: AI-generated or manipulated text that is published with the purpose of informing the public on matters of public interest must be disclosed as AI-generated. The Commission lists, among others, politics, public administration, justice, fundamental rights, public security, public health, environmental protection, consumer safety and economic, scientific or cultural developments that can be the subject of public debate. Exception: human review or editorial control where someone holds editorial responsibility. Mere spell-checking or grammar correction does not count as review. An AI text on such topics that an intern publishes without review: disclosure obligation. If the marketing lead reviews the substance of every post and takes responsibility: no obligation under the second subparagraph of paragraph 4. Whether a post covers a matter of public interest at all must be assessed case by case.
The exceptions nobody understands correctly
Three exception categories, three confusion traps.
Law enforcement (Art. 50(1) sentence 2, (3) sentence 2). For law enforcement there are exemptions authorised by law, subject to safeguards (Art. 50(1) and (3)). Systems available to the public to report a criminal offence remain covered by paragraph 1. If you work as an IT service provider for authorities, it is advisable to settle in the contract who carries which obligations.
Artistic/satirical content (Art. 50(4), first subparagraph). Reduces the duty, does not eliminate it. Disclosure in an appropriate manner that does not hamper the display or enjoyment of the work remains. Subtitle hint "AI-generated", credits entry, or a discreet watermark. What does not suffice: no hint at all.
Machine-readable is not enough. Paragraph 2 demands machine-readable marking (provider duty). But paragraph 4 demands "disclosure" for deepfakes, which under paragraph 5 must be provided "in a clear and distinguishable manner". According to the Commission, deployers cannot simply rely on the provider's marking (Art. 50 FAQ). A C2PA tag in the background XMP does not suffice if the end user cannot perceive it. Two layers: machine-readable marking and a notice people can perceive (visible or audible, depending on the medium).
What has applied since 02.08.2026, and what comes later
The deadline is constantly cited incorrectly. The facts:
- 02.02.2025: Prohibitions per Art. 5 (prohibited AI practices) and AI literacy duty per Art. 4 are active (Art. 113(a)).
- 02.08.2025: GPAI duties per Art. 53 ff. are active, as are the governance chapters and the penalties chapter (Art. 113(b)). More in our GPAI obligations guide.
- 02.08.2026: The AI Act applies in the main, including Art. 50. The AI Office and the national authorities begin enforcement. The Commission has launched an AI Act Complaints Tool, a Whistleblower Tool and a complaints channel for downstream providers using general-purpose AI models. For the marking obligation in Art. 50(2), systems already on the market before this date have until 02.12.2026.
- 02.12.2027: High-risk duties for Annex III systems (biometrics, critical infrastructure, education, employment, migration) apply only from this date. The Digital Omnibus on AI (Regulation (EU) 2026/1744, in force since 27.07.2026) moved them from 02.08.2026.
- 02.08.2028: High-risk duties for AI embedded in regulated products (Annex I) apply from this date, also after the Omnibus.
Meaning: since 02.08.2026 supervisory authorities can pursue breaches of Art. 50. According to the Commission, enforcement is mainly the job of the national market surveillance authorities. In Germany that is in principle the Bundesnetzagentur under the KI-MIG (in force since 29.07.2026), with exceptions such as BaFin for regulated financial activities. Details in our KI-MIG article.
Fines anchor
Art. 50 is on the list in Art. 99(4). That means: up to 15M EUR or 3% of worldwide annual turnover, whichever is higher. Not 35M (that is Art. 5). Not 7.5M (that is the tier for incorrect, incomplete or misleading information to authorities). It is a ceiling: the nature, gravity and duration of the infringement and the other circumstances of the case feed into the amount (Art. 99(7)). The Commission points out that proportionality can be taken into account for SMEs and small mid-cap companies.
Worked example: for a company that is neither an SME nor a small mid-cap and has 200M EUR turnover, the ceiling is 15M EUR, because that is the higher value (3% would be 6M EUR). For SMEs the lower of the two values applies under Art. 99(6), and for small mid-caps under paragraph 6a as well. For an SME with 20M EUR turnover the ceiling is therefore 3%, that is 600,000 EUR. Detailed math in our fines myth article.
5 actions if your UI is not yet adapted
The obligations have applied since 02.08.2026. If you have not started, work through this list now. Checking yourself against the 12 questions on the AI Act page helps as a starting point.
-
UI audit of all customer-facing AI systems. Who interacts directly with end customers in the product? Chatbots, voice IVR, recommendations, AI mail replies. List with owner and disclosure status.
-
Build the chatbot disclosure banner. Persistent badge in the conversation UI, German and English language versions, mobile tested. This is a frontend ticket, not a research project.
-
Marking plan for generative content. Who is provider in your setup? If you build an AI system yourself or offer it under your own name: review the Code of Practice (sign it or implement an equivalent), build the marking into the output pipeline, and note the 02.12.2026 deadline for legacy systems. If you white-label a vendor system: review contracts, ensure the marking passes through.
-
Deployer inventory with labelling status. Every AI system listed, classified (risk class), and its Art. 50 status (paragraphs 1, 3, 4). It fits straight into the 90-day audit readiness roadmap.
-
Review of all marketing materials. Who produced AI avatars, voice clones, generated images? What of that is still live? Content created before 02.08.2026 does not have to be labelled retroactively, but the Commission encourages it where possible. For new materials: check whether the deepfake or text labelling applies.
FAQ
If my UI already has a "powered by AI" footer, does that suffice? A footer alone is usually not "clear and distinguishable". Art. 50(5) demands the information at the latest at the time of the first interaction. Our advice: show it before that and keep it visible where possible.
We use ChatGPT Enterprise only internally. Does Art. 50 hit us? The obligation in paragraph 1 (chatbot notice) hits the provider, not you as a mere deployer. Paragraph 3 (emotion recognition) and paragraph 4 (deepfake, AI text) only hit you if you are in those use cases. Art. 4 (AI literacy) applies regardless, for purely internal use too.
Does a C2PA watermark suffice, or do I need both? Both, if you push deepfakes. The machine-readable marking is the provider duty from paragraph 2 (C2PA is one possible technique). The notice people can perceive (visible or audible, depending on the medium) is the deployer duty from paragraph 4 in conjunction with paragraph 5. According to the Commission, deployers cannot simply rely on the provider's marking.
What about existing customers who have used our AI since 2024? Since 02.08.2026 the obligations apply regardless of when the contract was signed. The only grace period concerns the marking obligation in paragraph 2 for systems placed on the market before 02.08.2026: it applies there only from 02.12.2026. Content created before 02.08.2026 does not have to be labelled retroactively.
Sources
- AI Act Regulation 2024/1689, Art. 50 (original): artificialintelligenceact.eu/article/50
- AI Act Service Desk EU Commission, Art. 50: ai-act-service-desk.ec.europa.eu
- European Commission, press release "Commission starts enforcing AI Act rules and new transparency requirements on 2 August" (31 July 2026): digital-strategy.ec.europa.eu
- European Commission, FAQ Transparency obligations under Article 50 (updated 24 July 2026): digital-strategy.ec.europa.eu
- European Commission, Code of Practice on Transparency of AI-generated Content (final 10 June 2026): digital-strategy.ec.europa.eu
- European Commission, "AI omnibus enters into force" (27 July 2026): digital-strategy.ec.europa.eu
- AI Act, Art. 5, Art. 99 (fines) and Art. 113 (application dates) as amended by the Omnibus: artificialintelligenceact.eu
- KI-MIG, BGBl. 2026 I No. 223 (German AI market surveillance act, unofficial translation of the title: Act on Market Surveillance and Innovation Promotion of Artificial Intelligence): gesetze-im-internet.de/ki-mig
- C2PA Standard (Coalition for Content Provenance and Authenticity): c2pa.org
Where you stand on Art. 50
The AI Act page walks you through the obligations with 12 questions, and the checklist arrives as a PDF by email. If you would rather clarify your situation in conversation: in a 30-minute call we go through it together.
Related reading: AI Act 90-day compliance plan, GPAI obligations August 2026, Fines myth, AI audit readiness 90 days, GDPR + Agentic AI Production.
EU AI Act
AI literacy under Article 4: build it and document it
Article 4 asks for measures that build AI literacy, but no specific level and no certificate. The checklist shows in 12 questions where your company stands.
About the author
Co-Founder · Business & Content Lead
Co-Founder of Sentient Dynamics. 15+ years of business strategy (incl. SAP), MBA. Writes about EU AI Act compliance, ROI measurement and how Mittelstand CTOs actually adopt agentic AI.