No. Article 50 of the EU AI Act does not require a visible label on every AI-made social post. It separates providers’ machine-readable marking of certain outputs from the visible disclosures that deployers must make in specific cases. Those rules apply according to the Act’s territorial scope, which can include some providers and deployers outside the EU when their AI output is used in the Union. The European Commission’s Article 50 guidance explains the two duties and their triggers.
A platform label and a legal disclosure do different jobs
Meta’s “AI info” label is a platform label. Meta says it may apply one when it detects industry-standard indicators of AI in an image or when someone discloses uploading AI-generated content. Meta also says its labeling method is evolving and may miss some AI-edited content. The label reflects that process; Article 50 uses separate legal tests. Meta describes its method.
That distinction matters when you see a label on a post. The label alone does not tell you which particular editing step prompted it, and Meta’s explanation does not make the label a legal definition of a deepfake.
Two duties: provider marking and deployer disclosure
A provider is responsible for a covered AI system; a deployer is the person or organization using the system and publishing or otherwise exposing its output. Article 50 assigns them different obligations.
| Actor | Article 50 duty | Trigger and scope | What the duty produces |
| Provider | Mark covered synthetic audio, images, video or text in an effective, machine-readable format that enables detection | Applies to providers of AI systems that generate synthetic content, subject to the Act’s scope and exceptions | A technical mark that systems can detect; it is not, by itself, a visible notice to viewers |
| Deployer | Clearly disclose qualifying deepfakes and specified AI-generated or manipulated public-interest text | Deepfakes must meet the legal test below. Text disclosure applies when there has been no substantive human review or editorial control with someone taking editorial responsibility | A disclosure visible or audible to people, no later than their first exposure to a qualifying deepfake |
The European Commission’s guidance describes exceptions to provider marking, including certain assistive functions for standard editing and some outputs not exposed to people. A provider’s technical mark does not take the place of a deployer’s disclosure when Article 50(4) requires one.
To work out which duty matters, start with the actor: the provider marks covered outputs for machine detection; the deployer discloses only when a defined legal trigger applies.
Which posts trigger a deployer’s disclosure duty?
For images, audio and video, the rule covers a deepfake: AI-generated or manipulated material that resembles existing people, objects, places, entities or events and would falsely appear authentic or truthful. A qualifying deepfake must be clearly disclosed by the deployer no later than a person’s first exposure. Artistic, creative, satirical and fictional works receive tailored disclosure treatment.
The separate text rule is narrower than a general requirement to label every AI-assisted caption. It applies when a deployer publishes AI-generated or manipulated text to inform the public about a matter of public interest. Disclosure is not required under this rule if the text has undergone human review or editorial control and someone takes editorial responsibility for it.
A natural person’s purely personal, non-professional use is excluded from the Act’s deployer obligations. The European Commission says regular economic benefit, or business, trade, occupational or freelance activity, counts as professional use. That personal-use exclusion concerns the deployer duty; it does not erase a provider’s separate obligation, and platform policies are a separate matter. The consolidated AI Act sets out the scope and definitions.
The Act can also reach some providers or deployers outside the EU when their AI system’s output is used in the Union. For a U.S.-based creator or provider, that EU connection—not location alone—is central to whether the Act’s territorial rules apply.
What the December 2, 2026 transition covers
Article 50 has applied since August 2, 2026. The later date is limited: providers of qualifying systems already placed on the market before August 2 have until December 2, 2026 to take the steps needed to meet Article 50(2)’s marking and detection duty. It is not a general extension for deployers’ disclosure duties. Regulation (EU) 2026/1744 sets out that transition.
What Meta’s label figures count
Meta’s reported figures also separate labeled content items from the total number of times labels appeared. For October 1–29, 2024, Meta reported the following on Facebook, Instagram and Threads. The figures are a dated company snapshot, not a count of unique people.
| Platform | Labeled items with label views | Labeled items with label actions | Aggregate label views | Aggregate label actions |
| More than 360 million | Nearly 6 million | More than 380 billion | More than 26 million | |
| More than 330 million | More than 13 million | More than 1 trillion | More than 60 million | |
| Threads | More than 730,000 | Nearly 70,000 | More than 150 million | More than 480,000 |
Meta defines a label view as the label rendering and a label action as an action such as clicking to learn more. The first two columns count labeled content items with those forms of activity; the last two report aggregate views and actions. Meta’s published figures and definitions distinguish these measures.
