EU AI labels are live—high-risk delays did not pause Article 50

The European Union’s Article 50 transparency obligations became applicable on August 2, 2026. Providers and professional deployers now have live duties covering direct AI interactions, machine-readable marking, deepfakes, emotion recognition, biometric categorisation and certain public-interest text; TechRadar’s August 2 account confirms that the separate delay affecting high-risk-system rules did not pause Article 50.

The rules apply to organizations operating in the EU and can reach providers established elsewhere when their systems’ output is used in the bloc. The August 2 start covers four distinct transparency tasks, while a four-month transition applies only to machine-readable marking by qualifying systems already on the market.

Article 50 assigns each disclosure to a specific role

The central distinction is between the provider that supplies an AI system and the deployer that uses one under its authority. A provider develops a system, has it developed or places it on the market under its own name or trademark. A deployer uses a system professionally; personal, non-professional activity is excluded, although regular commercial, occupational or freelance use can bring an individual within scope.

The European Commission’s Article 50 guidance defines the operative division, the exceptions and the limited transition:

• Provider of an interactive system: ensure that people are informed from the start of their first direct interaction that they are dealing with AI, unless this is already obvious to a reasonably well-informed and observant person.
• Provider of a generative system: make covered synthetic audio, images, video and text detectable through effective, reliable, robust and interoperable machine-readable marking.
• Deployer of emotion or biometric tools: inform people exposed to an emotion-recognition or biometric-categorisation system, whether the analysis occurs in real time or afterwards.
• Professional deployer publishing a deepfake: disclose the artificial generation or manipulation clearly and perceptibly by the audience’s first exposure.
• Deployer publishing covered public-interest text: label AI-generated or manipulated text unless it receives qualifying human review or editorial control and a person or entity holds editorial responsibility.

Employees acting under a company’s instructions are not separate deployers. The organization remains the deployer when contractors or freelancers operate the system on its behalf, under its responsibility and control.

Chatbot notices and technical marks solve different problems

The interaction rule addresses a person’s immediate understanding. It applies when an AI system conducts a genuine two-way exchange directly with a natural person, including a professional user. Background systems, machine-to-machine communication and systems that pass information through a human intermediary do not meet that direct-interaction test.

Machine-readable marking instead allows synthetic output to be detected technically. The provider duty covers generative systems, including general-purpose AI systems, producing synthetic audio, images, video or text. An embedded technical marker does not replace the visible or audible disclosure required when a professional deployer publishes a covered deepfake.

The marking obligation has defined limits. Excluded outputs include source code, short sequences of numbers, symbols or letters, machine-to-machine output never exposed to people and non-final material confined to closed industrial or product-development environments. Standard assistive editing is outside the duty, while a narrow exemption can apply to qualifying business-to-business or industrial uses.

The four-month transition does not postpone the whole article

The transition ends on December 2, 2026, and applies only to the Article 50(2) marking and detection duty for systems placed on the market before August 2. Chatbot notices, deployer disclosures and marking duties for systems newly placed on the market did not receive that extension.

Content generated before August 2 does not have to be labelled retroactively. Voluntary labelling remains possible, but it is not the same as the binding duties applying from the new start date.

The high-risk delay is legally separate. ITPro’s August 3 report confirms that stand-alone systems classified as high-risk under Annex III now face the relevant regime from December 2, 2027, while high-risk systems embedded in regulated products under Annex I move to August 2, 2028.

Deepfakes and public-interest text have tailored exceptions

A deepfake is AI-generated or manipulated image, audio or video content that resembles an existing or plausibly existing person, object, place, entity or event and can falsely appear authentic or truthful. Professional deployers must provide a clear, perceptible disclosure no later than first exposure.

For evidently artistic, creative, satirical, fictional or analogous works, the disclosure may be delivered in an appropriate way that does not hamper display or enjoyment. This changes how the notice can be presented; it does not create a blanket exemption from disclosure.

The text rule is narrower than a universal label for AI-assisted writing. It covers published text intended to inform the public about matters of public interest, including politics, public services, justice, fundamental rights, public security, public health, environmental protection, consumer safety and developments relevant to public debate.

Qualifying human review or editorial control removes the labelling duty when a person or entity also holds editorial responsibility for publication. The review must examine the substance using relevant knowledge and professional judgment, or give a responsible editorial authority power to approve, alter or reject it. Spell-checking, grammar correction and similarly superficial procedural checks are insufficient.

Enforcement has started, but the next deadline is narrowly drawn

National market-surveillance authorities will handle most enforcement. The European AI Office has a limited role for systems within its remit, while the European Data Protection Supervisor oversees relevant systems used by EU institutions, bodies and agencies.

For infringements of these obligations, fines can reach €15 million or, for an undertaking, 3% of its total worldwide annual turnover in the preceding financial year. Authorities must consider proportionality, including for small and medium-sized enterprises and small mid-cap companies, but their size does not create a general exemption.

The operative position is therefore split: Article 50 notices and disclosures have applied since August 2, while only qualifying pre-existing systems receive the marking transition to December 2, 2026. The later 2027 and 2028 dates govern the delayed high-risk requirements, not the transparency duties already in force.

Also read:

• Personal Loans For Bad Credit That Are Not Secured
• Where To Get The Best AI Article Writer
• Where Hybrid Workers Live Changes How Often They Stay Home