EU AI Act Article 50: What's Actually Enforceable Since August 2026 (and What Isn't Yet)

A deadline that's already passed, not a future one

EU AI Act Article 50 transparency and general-applicability obligations became enforceable on August 2, 2026 — meaning any organization whose AI systems reach EU users is already subject to these requirements, not preparing for a future compliance date. This distinguishes Article 50 sharply from the AI Act's high-risk Annex III obligations, which were delayed to December 2, 2027 under the AI Omnibus, and Annex I obligations, delayed further to August 2028 — a distinction that matters because conflating these timelines leads organizations to either over-invest in premature high-risk compliance work or under-invest in the transparency obligations that are already live.

What Article 50 actually requires, specifically

• AI interaction disclosure. Users must be clearly told when they're interacting with an AI system rather than a human, at the time of interaction — not buried in a terms-of-service document most users never read.
• Deepfake labeling. AI-generated or manipulated image, audio, or video content that could otherwise appear authentic must be labeled as AI-generated or manipulated.
• Machine-readable marking. Beyond a human-visible label, AI-generated content requires marking detectable by automated systems — a technical requirement distinct from simply adding a visible caption or watermark a person might notice.

The extraterritorial reach that catches non-EU companies off guard

The AI Act's extraterritorial scope catches any organization whose AI systems are used within the EU or produce outputs affecting EU residents, regardless of where the company itself is headquartered. US, UK, and APAC enterprises with any EU market exposure are equally subject to these rules — a common and consequential misunderstanding is assuming that being headquartered outside the EU exempts a company from Article 50 obligations.

First steps for identifying your actual obligations

The recommended starting sequence: identify your organization's actual AI applications, determine your specific role (provider, deployer, or both) for each one, rule out prohibited practices explicitly, and then implement the transparency and disclosure requirements that currently apply — rather than attempting a single blanket compliance exercise that treats every AI feature identically regardless of its actual role and risk category.

Third-party AI tools don't transfer away your obligation

Using a third-party AI chatbot, image generator, or other tool that itself lacks adequate transparency features doesn't relieve your organization of Article 50 obligations toward your own end users — the disclosure obligation runs to the people interacting with your product, regardless of which underlying AI vendor or model powers a given feature.

Some relief exists for smaller organizations — but it's not exemption

SME simplifications have been extended to small mid-caps under recent guidance, providing some genuine relief for smaller organizations navigating compliance — but this represents a reduced burden, not a full exemption from Article 50's core transparency obligations.

Treating this as ongoing, not a one-time August 2026 exercise

A new AI-powered feature launched after your initial compliance review can introduce new Article 50 obligations that a one-time assessment won't catch. Organizations that hold up well treat AI transparency compliance as a standing part of new feature review, not a single project with a defined end date tied to the original enforcement deadline.