Art.50

Article 50 — Transparency Obligations for Providers and Deployers of Certain AI Systems

Transparency SME Relevant ~3 min read · 603 words

Article 50 opens Chapter IV and introduces transparency obligations that apply beyond high-risk AI — covering any AI system that interacts with people, generates synthetic content, or makes automated decisions. Chatbots must disclose they are AI. Deepfakes must be labelled. Emotion recognition systems must inform users. These are among the most visible and broadly applicable obligations in the entire Act.

! High compliance impact for SMEs

WHAT THE ARTICLE IS ABOUT

Disclosure obligations for AI systems that interact with or affect people

Article 50 establishes transparency obligations for a broad range of AI systems outside the high-risk framework. It recognises that many AI systems — chatbots, emotion recognition systems, deepfake generators, content recommendation engines — interact directly with people in ways that require disclosure, even if they are not classified as high-risk. These obligations apply to both providers and deployers.

WHAT IT SAYS

Four categories of AI systems with disclosure obligations

Category 1 — Chatbots and conversational AI: Providers must ensure that AI systems interacting with humans are designed to inform users they are communicating with an AI — unless this is obvious from context. The disclosure must happen at the latest at the start of the first interaction
Category 2 — Emotion recognition and biometric categorisation: Deployers must inform individuals when they are subject to an emotion recognition system or biometric categorisation system
Category 3 — Deepfakes and synthetic media: Deployers of AI systems generating or manipulating image, audio or video content that could be mistaken for real content must disclose the artificial nature of that content — this applies to deepfakes, AI-generated images, synthetic voices, and similar
Category 4 — AI-generated text on public interest matters: Where AI generates text about matters of public interest and that text is published, it must be disclosed as AI-generated — with exceptions for authorised use in detecting illegal content or where human editorial review has taken place
All disclosures must be provided in a clear and distinguishable manner at the latest at the time of first interaction or exposure
These obligations do not apply where the AI use is authorised by law for law enforcement detection of criminal offences, or where the artificial nature is evident from context

WHO IS AFFECTED

A wide range of providers and deployers beyond the high-risk framework

Providers of chatbots, virtual assistants, customer service AI, and any conversational AI system
Deployers using emotion recognition or biometric categorisation AI in customer-facing or workplace contexts
Providers and deployers of AI image, audio and video generation tools — including marketing content, synthetic voiceovers, and AI-generated imagery
Publishers and content platforms deploying AI to generate text on public interest topics
Social media platforms and content recommendation services using AI that may generate or manipulate content

WHAT IT MEANS FOR SMES

The most broadly applicable article in the Act for everyday AI use

This is the article most likely to directly affect SMEs who are not in the high-risk AI space — if you use a chatbot on your website, generate AI images for marketing, or use synthetic voiceovers, you have obligations under Article 50
Chatbot disclosure is simple to implement: a clear statement at the start of interactions that the user is talking to an AI, or a visible label in the interface
Deepfake and synthetic media labelling applies even to clearly benign uses — AI-generated marketing images, synthetic spokesperson videos, and similar content must be disclosed as AI-generated
The emotion recognition disclosure is particularly relevant for HR technology — if you use any tool that claims to analyse facial expressions, voice tone, or emotional states in interviews or workplace settings, you must inform the people being assessed
The obligations come into effect from August 2026 — check whether your current AI deployments already comply with these disclosure requirements

Article 5 — Prohibited AI practices (emotion recognition in workplaces is prohibited in certain contexts — check Article 5 before relying solely on Article 50 disclosure)
Article 13 — Transparency for high-risk AI (the high-risk transparency obligation — Article 50 covers the non-high-risk space)
Article 3 — Definitions (defines emotion recognition system, deepfake, and other relevant terms)
Article 26 — Deployer obligations (for high-risk AI deployers, transparency obligations are covered there instead)

← Back to the Guide