EU AI Act · Article 50 · Regulation (EU) 2024/1689
A label says what happened. The workflow proves it.
Article 50 of the EU AI Act applies from August 2, 2026 and sets transparency obligations for providers and deployers of certain AI systems. Depending on the system and use case, these can include informing people 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. A limited transition until December 2, 2026 applies only to the marking and detection obligation under Article 50(2) for certain systems already on the market before August 2.
Are we the provider, the deployer, or both under Article 50?
It depends on the workflow. In plain terms, a provider develops an AI system or places it on the market under its own name, while a deployer uses an AI system under its authority in the course of professional activity. The same organization can be both across different steps in the same content chain. The practical starting point is to map each workflow step, identify which Article 50 duty applies there, and keep a short record of who owned the decision, which output was affected, and what evidence still survives after handoff. Numbers Protocol supports that evidence layer. This page is not legal advice.
A label communicates a conclusion. An audit trail preserves the evidence behind it.
After content is exported, emailed, uploaded, converted, and compressed by a platform, the visible label may survive. The supporting evidence often does not. An organization may no longer be able to show which source asset was used, what the AI changed, which person substantively reviewed which version, or whether the evidence can still be inspected by a third party.
The one-asset test
Take one real AI-assisted asset and follow it through your existing workflow. It should preserve five elements:
Content identity
A durable reference that distinguishes the asset from earlier drafts, copies, and derivatives.
AI-use decision
A record of whether AI was used and which disclosure or labelling decision was made.
Provenance or machine-readable evidence
Inspectable information for a verifier, not only a screenshot or a written claim.
Accountable reviewer
The responsible reviewer and reviewed version remain identifiable.
Downstream verification
The evidence remains accessible after export, publication, transfer, or archiving.
Choose the situation closest to your workflow
Compare the six situations, then choose the one most relevant to your current work.
Newsrooms publishing AI-assisted text
Preserve who reviewed which version, what was decided, and evidence that outlives the original CMS.
Platforms and creators handling deepfakes
Preserve evidence of what was generated or manipulated after platform re-encoding.
Providers marking AI output at the source
Keep machine-readable marking intact and verifiable after export.
Teams running autonomous AI agents
Tie every agent output to a durable, inspectable action receipt.
Anyone whose content leaves the original tool
Keep provenance with the file so third parties can verify it without the original system.
Creators proving their work is human-made
Record when and how a person made the work from the moment of creation.
Not a compliance badge. The record layer behind the label.
Numbers Protocol does not automatically make an AI system compliant. It supports the evidence layer: Numbers ID (NID) for persistent references, C2PA-compatible provenance that travels with content, and Capture SDK, ProofSnap, and Capture Cam for inspectable receipts in product, platform, and AI-agent workflows.
Review your Article 50 workflow with Numbers Protocol
Legal sources: European Commission Article 50 implementation guidelines; Code of Practice on Transparency of AI-Generated Content; Regulation (EU) 2024/1689. This page is provided for general informational purposes and does not constitute legal or compliance advice.
EU AI Act · 第 50 條 · Regulation (EU) 2024/1689
標籤說明發生了什麼。流程負責證明。
2026 年 8 月 2 日起，EU AI Act 第 50 條針對特定 AI 系統的提供者與部署者引入透明度義務，依情境可能包括：告知使用者正在與 AI 互動、為 AI 生成或操縱的內容加上機器可讀標記，以及揭露特定深偽內容或 AI 生成的公共利益文字。至 2026 年 12 月 2 日的有限過渡期，僅適用於 8 月 2 日前已投放市場之特定系統在第 50(2) 條下的標記與偵測義務。
我們在第 50 條下是提供者、部署者，或兩者皆是？
視工作流程而定。簡單來說，提供者開發 AI 系統，或以自己的名稱將其投放市場；部署者則在專業活動中，於其權限下使用 AI 系統。同一組織在同一內容鏈的不同步驟，可能同時扮演兩種角色。實務起點是逐步繪製工作流程、識別各步驟適用的第 50 條義務，並保留一份簡短紀錄，說明誰負責決策、哪個輸出受到影響，以及哪些證據在交接後仍然存在。Numbers Protocol 支援這個證據層。本頁不構成法律意見。
標籤傳達一個結論。稽核紀錄保存結論背後的證據。
內容經過匯出、寄送、上傳、轉檔與平台壓縮後，可見的標籤也許還在，但支持它的證據往往已經不見。組織可能無法再證明使用了哪個來源素材、AI 修改了什麼、哪位審核者實質審核了哪個版本，以及第三方是否仍能檢驗證據。
一資產測試
挑一件真實的 AI 協作資產，走一遍現有流程，並保留五個要素：內容識別、AI 使用決定、出處或機器可讀證據、可歸責的審核者、下游可驗證。任何一項在第一次交接就消失，流程就是脆弱的。
選擇最接近你的工作情境
先比較六種情境，再選擇目前與你的工作最相關的一種。
新聞編輯室發布 AI 輔助文字
保留誰審核了哪個版本、做了什麼決定，以及能離開原始 CMS 的證據。
平台與創作者處理深偽內容
在平台重新壓縮轉檔之後，仍保留生成或操縱內容的證據。
提供者在源頭標記 AI 產出
讓機器可讀標記在匯出後仍保持完整且可驗證。
團隊運行自主 AI agent
讓每一個 agent 產出對應持久、可檢驗的行動收據。
內容會離開原本工具的每一個人
讓出處紀錄跟著檔案走，使第三方不需原始系統也能驗證。
創作者證明作品出自人手
從創作當下留下何時、如何由人完成作品的紀錄。
不是合規徽章，是標籤背後的紀錄層。
Numbers Protocol 不會自動讓 AI 系統合規。它支援證據層：Numbers ID (NID) 提供持久參照、C2PA 相容的出處紀錄隨內容移動，以及 Capture SDK、ProofSnap 與 Capture Cam 為工作流程留下可檢驗的收據。
與 Numbers Protocol 檢視你的 Article 50 工作流程
法規依據：歐盟執委會 Article 50 實施指引；AI 生成內容透明度實務守則；歐盟法規 2024/1689。本頁僅供一般資訊參考，不構成法律或合規意見。