EU AI Act and AI-generated content: what must be labelled from 2 August 2026? - WebsiteInit
From 2 August 2026, the EU AI Act requires disclosure of AI-generated or substantially altered content only in specific contexts: realistic synthetic media (deepfakes), direct AI system interactions (chatbots), and AI-generated text addressing public-interest topics (politics, health, safety, environment) without substantive human review and editorial responsibility.
Routine product descriptions, ordinary advertisements, and private correspondence do not require AI labels; disclosure is mandatory only when content could mislead audiences about authenticity or inform them on matters of public debate without competent fact-checking and human accountability for publication.
Providers must design conversational AI systems to disclose their artificial nature at the start of interaction and add machine-readable marking to generated content where required; deployers (businesses using the systems) are responsible for ensuring visible disclosure of deepfakes and public-interest text and for maintaining editorial oversight.
The substantive-review exception applies exclusively to public-interest text and requires both a competent person's deliberate examination of facts, sources, and meaning with genuine editorial control, and organizational acceptance of responsibility for publication; this exception does not apply to deepfakes or direct AI interactions.
Organizations must assess each piece of content individually based on its subject, purpose, and audience rather than applying blanket rules; AI literacy requirements, combined with disclosure obligations and human-review standards, establish a framework that prioritizes transparency and accountability over automated content generation.