The AI Act delay did not stop Article 50

The EU delayed part of the AI Act. It did not delay the duty to tell a person they are talking to an AI system.

That distinction went live on August 2, 2026. Regulation (EU) 2026/1744 moved the main high-risk-system requirements to December 2, 2027 for Annex III systems and August 2, 2028 for systems tied to Annex I product law. But the consolidated AI Act still applies Chapter IV, including Article 50, from August 2, 2026.

So “the AI Act was delayed” is not a compliance status. It is a headline. A customer-facing assistant can be outside the delayed high-risk timetable and already inside a live transparency duty.

One law, two clocks

The Digital Omnibus amendment is specific about what moved: Sections 1, 2 and 3 of Chapter III for high-risk AI systems. Article 50 sits in Chapter IV. The amendment also added a narrow four-month transition for one Article 50 duty, which is another clue that the whole article did not move.

Article 50 is not one label rule either. It assigns different work to different actors:

Product surface	Duty	Actor named in the Act
AI system interacting directly with a person	Inform them they are interacting with AI, unless it is obvious	Provider, Article 50(1)
System generating synthetic audio, image, video or text	Make outputs machine-readable and detectable as AI-generated or manipulated	Provider, Article 50(2)
Emotion recognition or biometric categorisation	Inform the people exposed to it	Deployer, Article 50(3)
Deepfakes and certain public-interest text	Disclose the AI generation or manipulation, subject to stated exceptions	Deployer, Article 50(4)

That provider/deployer split matters more than the press release. Under Article 3, a provider is the entity that develops a system, or has it developed, and puts it on the market or into service under its own name or trademark. A deployer uses a system under its authority. Buying a third-party model does not automatically settle which role your finished product occupies.

The grace period is smaller than the headline

There is a transition, but only for Article 50(2). Providers of synthetic-content systems placed on the market before August 2, 2026 have until December 2, 2026 to meet the machine-readable marking and detection duty. The Commission's Article 50 FAQ states that the grace period does not extend to the other transparency obligations.

That means a legacy chatbot does not inherit a four-month holiday for the Article 50(1) interaction disclosure. Nor does a deployer get to replace a visible deepfake disclosure with the provider's invisible machine-readable mark. The Commission says those are separate duties.

There are real scope details and exceptions here: the obviousness test for direct interaction, assistive editing, law-enforcement uses, artistic works, and human editorial control over public-interest text. This is an engineering map, not a substitute for counsel on a particular product. But “wait until 2027” is not a safe default.

Put the role beside the deadline

My rule is to ban a single AI Act status field from the launch checklist. Record four things instead: the product surface, your role for that system, the exact paragraph, and the effective date. A green status without those four values hides the decision you will need to defend.

This extends a practice I described in the US compliance-date piece: stamp the legal version and assessment date into the feature documentation. For Article 50, add the role. That one column stops a team from assigning a provider duty to every buyer, or missing that its own branded system may make it a provider.

Run the check at the product-surface level. “We use generative AI” is too broad. “Our support assistant talks directly to EU customers; we put the system into service under our name; Article 50(1) applies from August 2” is the kind of statement an engineer can implement and a reviewer can challenge.

A delayed chapter is not a delayed law—map each AI surface to its role, paragraph and date before a headline moves your roadmap.