EU AI Act Article 50 Now Applies: What AI-Generated Content Transparency Means in 2026
Published September 2026 by Solarc Labs
Article 50 has applied since 2 August 2026
The European Commission’s final guidelines state that Article 50 transparency obligations apply from 2 August 2026. They cover several different situations, including informing people when they interact directly with certain AI systems, machine-readable marking of AI-generated or manipulated content by relevant providers, and disclosure duties for deployers in defined cases such as deepfakes and certain public-interest text. The Commission FAQ also describes a limited transition for the marking-and-detection obligation for certain AI systems placed on the market before 2 August 2026, with compliance for that obligation from 2 December 2026. Do not reduce all of those obligations to one generic “AI label” requirement. The first task is to identify whether the organisation is acting as a provider or deployer, which content or interaction is in scope, whether a transition applies and which obligation actually applies.
Machine-readable marking and visible disclosure are related but not the same evidence
The Commission distinguishes provider duties around machine-readable marking from deployer duties to inform people in specified situations. A content operation may therefore need more than one control: evidence about available provenance or marking, plus a human-reviewed decision about visible disclosure at publication time. Keeping these controls separate also prevents a common mistake: assuming that the absence of detectable provenance proves a file was human-created. Lack of a manifest or mark is an unknown evidence state, not authorship proof.
C2PA can support provenance review, but it is not an AI detector or legal verdict
Content provenance standards can provide machine-readable evidence about the history or assertions attached to a supported asset. That can be useful in a transparency workflow, but provenance inspection does not answer every Article 50 legal question and it cannot reliably infer human authorship when no provenance data is present. A defensible workflow keeps the original asset, records what provenance evidence was actually found, preserves an explicit unknown state, adds human generation or manipulation context and records the publication or disclosure decision separately.
Build a publication checklist around the real role and content type
Before publication, record who is providing or deploying the AI system, the type of content, whether the organisation knows it was generated or manipulated with AI, what machine-readable evidence is present, whether a visible disclosure requirement may apply and who owns the final decision. AI Transparency Pack supports the evidence side of this workflow for a bounded batch of supported media using a pinned C2PA toolchain plus human disclosure context. It does not generate content, remove watermarks, detect AI authorship or certify compliance with the AI Act.
Primary sources
Sources used for this article
Continue the job