EU AI Act Article 50 — Transparency obligations

Note: this explainer is informational and framework references are never a compliance guarantee — align your program with your legal counsel.

The EU AI Act transparency obligations in Article 50 apply regardless of risk class: even a minimal-risk system has to be honest about the fact that it is AI, and about content it generates. Under Article 113 of the Regulation these duties apply from the general date of 2 August 2026 — the nearest AI Act deadline most operators face. A single narrow exception (below) gives providers of generative AI already on the market until 2 December 2026 for the machine-readable marking duty.

Chatbots must disclose they are AI

Article 50(1): a system that interacts directly with people must make clear they are dealing with an AI, unless that is already obvious from the context. A support bot, a website assistant, or an AI voice line all fall here — the disclosure has to reach the person before or as they start interacting.

AI-generated content and deepfakes must be marked

Two related duties:

• Synthetic content marking (50(2)) — providers of generative AI must mark audio, image, video, or text output as artificially generated or manipulated, in a machine-readable format that downstream systems can detect. This applies from 2 August 2026 for systems placed on the market from that date; providers of generative AI systems already on the EU market before 2 August 2026 have a transitional period until 2 December 2026 to bring the marking into compliance — a transitional period under Articles 50 and 113 of the Regulation itself, not the Digital Omnibus.
• Deepfake disclosure (50(4)) — deployers that use AI to generate or manipulate image, audio, or video that resembles real people, places, or events (a deepfake) must disclose that it is artificially generated. AI-generated text published to inform the public on matters of public interest carries a similar disclosure duty.

For operators: which of your tools trigger this

If your teams use generative AI to produce customer-facing copy, images, audio, or video — or run any assistant that talks to customers — Article 50 is already in scope for those workflows. The first control is knowing where that use is happening, because an undisclosed AI chatbot or an unmarked AI image is only a risk if you can't see that it exists.

Shield's lever: Discovery surfaces the generative-AI tools employees actually use, and the audit log keeps the record of that use — so the transparency question becomes "which of these known tools produces public-facing output?" rather than a blind guess.

The outcome

By 2 August 2026 the Article 50 duties reward the operators who already know their generative-AI footprint: which assistants face customers, which tools produce published content, and who owns each. Applying the EU AI Act & GDPR Starter bundle from the preset library builds that footprint today, so transparency becomes a labelling decision on a known list — not a hunt.