# EU AI Act Article 50: The Label Is Not the Audit Trail > Article 50 of the EU AI Act applies from August 2, 2026 and creates transparency obligations for providers and deployers of certain AI systems. Depending on the system and use case, these obligations can include informing people that they are interacting with AI, adding machine-readable markings to AI-generated or manipulated content, and disclosing certain deepfakes or AI-generated public-interest text. Canonical page: https://numbersprotocol.io/eu-ai-act-article-50 Last updated: 2026-08-04 ## What changes on August 2, 2026? Article 50 transparency obligations start to apply on August 2, 2026. They can include: 1. Informing people when they are interacting directly with an AI system. 2. Marking AI-generated or manipulated content in a machine-readable format so it can be detected. 3. Disclosing certain deepfake image, audio, or video content. 4. Disclosing certain AI-generated or manipulated text published to inform the public on matters of public interest. The exact obligation depends on the system, role, and use case. The exception for public-interest text requires substantive human review and editorial responsibility. Formal checks such as spelling or grammar correction are not enough by themselves. ## Is every Article 50 obligation delayed until December 2, 2026? No. The limited transition until December 2, 2026 applies only to the marking and detection obligation under Article 50(2) for certain systems placed on the market before August 2, 2026. It is not a general delay for every Article 50 transparency obligation. ## Why is a visible AI label not enough? A label tells an audience that AI was involved. It does not automatically preserve the evidence behind that statement. After content is exported, emailed, uploaded, converted, or compressed by a platform, an organization may no longer be able to show: - Which source asset or input was used. - What the AI generated or changed. - Which disclosure or marking decision was made. - Who substantively reviewed which version. - Whether a third party can still inspect the evidence after publication. The label communicates a conclusion. An audit trail preserves the evidence supporting that conclusion. ## What is the one-asset test? Take one real AI-assisted asset through the complete workflow: generation, recording, export, substantive review, disclosure decision, publication, and independent verification. Check whether these five elements survive: 1. **Content identity:** A persistent reference distinguishes the asset from drafts, copies, and derivatives. 2. **AI-use decision:** The workflow records whether AI was used and which disclosure or marking decision was made. 3. **Provenance or machine-readable evidence:** A verifier can inspect information, not just a screenshot or written claim. 4. **Accountable reviewer:** The responsible reviewer and reviewed version remain identifiable. 5. **Downstream verification:** The evidence remains available after export, publication, transfer, or archiving. If any element disappears at the first handoff, the workflow is fragile. ## Which workflows should test Article 50 evidence? ### Newsrooms publishing AI-assisted text Preserve who reviewed which version, what decision was made, and evidence that survives outside the original content management system. ### Platforms and creators handling deepfakes Preserve evidence of what was generated or manipulated, including after a platform recompresses or converts the media. ### Providers marking AI output at the source Check that machine-readable markings survive export and remain verifiable downstream. ### Teams running autonomous AI agents Keep persistent, inspectable action receipts for agent outputs, modifications, publication, and other actions. ### Workflows where content leaves the original tool Keep provenance records with the file so a third party can verify the content without access to the original system. ### Creators proving work was human-made Record when and how a work was created by a person so others can verify its human origin. ## Where does Numbers Protocol fit? Numbers Protocol does not determine whether a specific organization falls within Article 50, replace legal advice, or automatically make an AI system compliant. It supports an inspectable evidence layer: - **Persistent identity:** Numbers ID (NID) gives each asset or record a durable reference. - **Portable provenance:** C2PA-compatible provenance can travel with content instead of remaining inside the original tool. - **Action and review receipts:** Capture SDK, ProofSnap, and Capture Cam can add inspectable records to product, platform, AI-agent, and media workflows. To discuss one of these workflows, use the [Numbers Protocol contact form](https://numbersprotocol.io/contact-us?source=article50). ## Official sources - [European Commission guidelines for Article 50 transparency obligations](https://digital-strategy.ec.europa.eu/en/library/guidelines-transparency-obligations-providers-and-deployers-ai-systems) - [European Commission Code of Practice on Transparency of AI-Generated Content](https://digital-strategy.ec.europa.eu/en/policies/code-practice-ai-generated-content) - [Regulation (EU) 2024/1689](https://eur-lex.europa.eu/eli/reg/2024/1689/oj/eng) This guide is provided for general informational purposes and does not constitute legal or compliance advice.