NewsBench: Expert-Grounded Evaluation of Epistemic Quality in AI News Reporting

arXiv:2610.33240v1 Announce Type: new Abstract: As frontier AI systems increasingly serve as sources of political and current-events information, evaluating the epistemic quality of their responses is now especially consequential. Existing evaluations often focus narrowly on ideological alignment, rely on measures not validated for this setting, or depend on reference labels derived from public opinion or non-expert annotation, potentially producing misleading or miscalibrated results. To address these limitations, we introduce NewsBench, an expert-grounded benchmark that reframes evaluation around editorial standards developed by senior practitioners in journalism, policy, and intelligence analysis. NewsBench operationalizes quality across three dimensions, source quality, factuality, and neutrality, using a two-phase pipeline in which domain experts first define and calibrate evaluation guidelines, which are then transferred to automated AI judges validated against expert-adjudicated gold labels. The benchmark includes approximately $3,000$ evaluation prompts, with roughly $500$ refreshed monthly to capture emerging current events, roughly $2,500$ human expert-annotated items for calibration and validation, and a suite of AI judges designed to enable scalable evaluation. To validate our methodology, we report both inter-rater reliability among human experts and AI judge--human agreement across all three dimensions. Finally, we apply NewsBench to four frontier models, revealing distinct and non-overlapping failure profiles in sourcing, factuality, and neutrality.

NewsBench reveals failure profiles in AI news reporting

The introduction of an expert-grounded benchmark for AI news reporting quality helps address concerns about misinformation and bias by enabling reliable evaluation of epistemic quality. However, it also reveals notable failure profiles in leading AI models, indicating persistent risks in AI-generated political and current-events content. Overall, it offers cautious optimism with significant challenges remaining.

This is a benchmarking paper, not a claim about catastrophic or utopian outcomes. It documents that frontier models have distinct weaknesses in news-related epistemic quality, a mildly concerning but well-evidenced finding, balanced by the constructive value of the evaluation tool itself.

This paper introduces a new benchmark, NewsBench, for evaluating the epistemic quality of AI systems used for news. This work is positive because it aims to improve the reliability and trustworthiness of AI-generated information, which can mitigate risks associated with misinformation. It is a research proposal but describes an implemented system and reports initial validation.

The paper introduces a benchmark for evaluating AI news reporting based on expert standards. This could lead to more reliable and trustworthy AI-generated news. The methodology is well-defined and validated, indicating a strong potential for positive impact.

SOURCE     arXiv cs.AI (tier 1)
   ↓
DOCUMENT   07297e3d-d572-48d0-8609-5e527f8eb40e
           https://arxiv.org/abs/2610.33240
   ↓
EVIDENCE   3 extracted excerpts
   ↓
MODEL RUN  5 runs, methodology 1.0.1
   ↓
SCORE      -34  (Adverse)
   ↓
CONFIDENCE 74%