What are transparency obligations?
Transparency obligations are the EU AI Act’s requirements that people be told when AI is involved and that the companies using AI systems receive enough information to operate them properly. The Act uses the term in three places. Article 50 sets disclosure duties that apply to specific kinds of systems whatever their risk tier. Article 13 requires providers of high-risk AI systems to design them so deployers can understand and use the output, and to supply instructions for use. The chapter on general-purpose AI models requires documentation for downstream providers and a public summary of training content. Article 50 is the part most companies meet first, because it covers ordinary chatbots and content generation.
How transparency obligations work
Under Article 50, a provider of a system that interacts directly with people must design it so that people know they are dealing with AI, unless that is obvious from the context. A provider of a system that generates synthetic audio, image, video, or text must mark the output as artificially generated in a machine-readable format, which is where AI content watermarking and content provenance standards come in. Deployers have their own duties. They must disclose deepfakes, meaning generated or manipulated content that resembles real people, places, or events. They must disclose AI-generated text published to inform the public on matters of public interest, unless a human has reviewed it and someone holds editorial responsibility. They must also inform people exposed to emotion recognition or biometric categorization systems. The disclosure has to be clear and given no later than the first interaction.
Transparency obligations in practice
An online retailer runs a support chatbot and uses an image model to produce lifestyle photos for its catalog. The chatbot opens every conversation with a sentence saying that the customer is talking to an automated assistant and how to reach a person. The image pipeline keeps the generator’s embedded watermark and adds provenance metadata, and the marketing policy states that any generated image showing a recognizable real place or person as if real is labeled on the page. Employees who use a drafting assistant for blog posts review every post before publication, which keeps those posts outside the public-interest disclosure rule.
What to watch for
- The duty to disclose a chatbot sits with the provider by design and with the deployer in practice, so check who configures the greeting.
- Stripping a vendor’s watermark from generated content to make it look cleaner can create a breach.
- A company that builds its own high-risk system is both provider and deployer, and still has to write the instructions for use.
- Treat disclosure as a minimum, because a customer who discovers the AI on their own trusts the company less than one who was told.