EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Evaluating generative AI safety on real-world harm: A focused classification study using the RealHarm dataset

Matthew Clark, Lucía Martín, Javier Moreno

AI and Wisdom · 01.06.2026 · 1 (2)

DOI: 10.68710/9zak8c ↗

Kurz gefasst

Fachartikel in der Zeitschrift AI and Wisdom, online erschienen am 01.06.2026. Verfasst von Matthew Clark und 2 weiteren Autor:innen. Die Arbeit nennt 14 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Safety benchmarks for large language models often rely on synthetic prompts, adversarial test cases, or narrowly defined toxicity tasks, which can leave a gap between benchmark performance and the kinds of failures that appear in deployed systems. This study evaluates whether a general-purpose large language model can distinguish safe from unsafe AI assistant interactions using RealHarm, a benchmark derived from publicly reported real-world failures and paired with human-reviewed safe rewrites. The full sample contained 136 observations, consisting of 68 unsafe interactions and 68 matched safe counterparts across 11 primary harm taxonomies. All records were anonymized, randomly shuffled, and classified in a blind single-pass procedure using the same binary SAFE/UNSAFE instruction. The evaluated model achieved 91.2% accuracy (95% Wilson CI 85.2% to 94.9%), 100.0% precision, 82.3% recall, 90.3% F1, and 100.0% specificity. All 12 classification errors were false negatives; no safe interaction was incorrectly flagged as unsafe. Errors were concentrated in misinformation and unsettling multi-turn interactions, with additional misses involving privacy, vulnerable users, interaction breakdowns, and brand-damaging conduct. Qualitative inspection showed that many missed cases were superficially fluent and helpful, depended on external factual knowledge, emerged late in long conversations, or involved subtle role and contextual boundaries. The findings suggest that aggregate safety accuracy can conceal a consequential asymmetry between avoiding over-restriction and detecting subtle harm. For governance and deployment, the results support category-aware monitoring, explicit fact-verification mechanisms, and multi-stage review for longer or context-sensitive interactions.

Zitieren

APA

Clark, M., Martín, L., & Moreno, J. (2026). Evaluating generative AI safety on real-world harm: A focused classification study using the RealHarm dataset. AI and Wisdom, 1(2). https://doi.org/10.68710/9zak8c

Vancouver

Clark M, Martín L, Moreno J. Evaluating generative AI safety on real-world harm: A focused classification study using the RealHarm dataset. AI and Wisdom. 2026;1(2). doi:10.68710/9zak8c

BibTeX
@article{Clark2026,
  author = {Clark, Matthew and Martín, Lucía and Moreno, Javier},
  title = {Evaluating generative AI safety on real-world harm: A focused classification study using the RealHarm dataset},
  journal = {AI and Wisdom},
  year = {2026},
  volume = {1},
  number = {2},
  doi = {10.68710/9zak8c}
}
Kostenlos starten →