Transparency obligation
A legal requirement to tell people when they are talking to or being assessed by an AI system rather than a human.
The Simple Version
A legal requirement to tell people when they are talking to or being assessed by an AI system rather than a human.
Detailed Explanation
Article 50 applies to AI systems intended to interact directly with natural persons (e.g. chatbots), systems generating synthetic audio, image, video, or text content, and emotion recognition or biometric categorisation systems. Providers must design systems to automatically inform users they are interacting with AI unless this is obvious. Deployers using emotion-recognition or biometric categorisation AI must inform persons exposed to those systems. Deep-fake content must be labelled as artificially generated. These obligations apply to limited-risk AI systems and reinforce the Act's commitment to human autonomy.
Key Characteristics
- Applies to chatbots, synthetic media, emotion recognition, and biometric categorisation
- Disclosure must be machine-readable where technically feasible for synthetic content
- Obligation falls on both providers (by design) and deployers (in operation)
- Exceptions apply where context makes AI involvement evident
Why It Matters
Customer-service chatbot operators, marketing content teams producing AI-generated imagery, and HR platforms using emotion analysis must update user journeys and disclosures to satisfy Article 50.
Real-World Analogy
Like food-labelling rules that require allergen declarations regardless of whether consumers ask, transparency obligations require AI disclosure proactively rather than only when users inquire.
Common Misconceptions
- Transparency obligations only apply to high-risk AI, they apply to a broader set of interactive and synthetic-media AI systems at the limited-risk tier.
- Disclosing at sign-up is sufficient, the disclosure must be made at the point of each interaction or content exposure.