Transparency Obligations

Some AI is not dangerous so much as potentially deceptive. The Act’s answer for the limited-risk tier is light: do not let people be fooled about whether they are dealing with a machine or with synthetic content. These rules apply broadly - most product teams will owe at least one of them.

The Idea: Disclosure, Not Restriction

Limited-risk systems are allowed to do what they do. The only requirement is honesty about their nature. The transparency rules sit in their own article and apply to both providers and deployers depending on the case. They become applicable in August 2026, alongside the bulk of the Act.

The Four Disclosure Duties

1. "You are talking to an AI"

Providers must design AI systems intended to interact directly with people - chatbots, voice assistants, virtual agents - so that those people are informed they are interacting with an AI, unless it is obvious from the context to a reasonably well-informed person. The disclosure must be clear and come at the first point of interaction.

2. Label AI-generated and manipulated content (deepfakes)

Deployers who use an AI system to generate or manipulate image, audio, or video content that constitutes a deepfake must disclose that the content has been artificially generated or manipulated. There are tailored carve-outs - for example for evidently artistic or satirical work, where the disclosure is provided in a way that does not spoil the work.

3. Mark synthetic content as machine-readable

Providers of generative AI systems must ensure their outputs are marked in a machine-readable format and detectable as artificially generated or manipulated - the technical underpinning (watermarking, metadata, provenance signals) that lets platforms and tools identify synthetic media at scale, not just human readers.

4. Disclose AI-written text on matters of public interest

Deployers who use AI to generate or manipulate text published to inform the public on matters of public interest must disclose that it is AI-generated - unless the content has undergone human review and a person or organisation holds editorial responsibility for it.

What "Good" Disclosure Looks Like

• Up front and unmissable. A chatbot should announce its nature before the conversation, not bury it in a privacy policy.
• Plain language. "You’re chatting with an automated assistant" beats legalese.
• Both human-facing and machine-readable for generated media. A visible label for people and embedded provenance metadata for platforms.
• Accessible. Disclosure should work for users with assistive technology, not just sighted readers.

Transparency Is a Floor, Not a Ceiling

Meeting the limited-risk duties does not exempt you from anything else. A customer-service chatbot owes the disclosure duty and still must respect data-protection law; a generative system owes content marking and, if the model is GPAI, the duties from Lesson 5. Transparency is the minimum honesty requirement, layered on top of whatever else applies.