LATIDIA · Ciberseguridad

Generando riesgos: Evaluación comparativa y reducción de riesgos para la seguridad alimentaria en modelos de lenguaje a gran escala.

arXiv:2604.01444v3 Anuncio Tipo: reemplazar Resumen: La seguridad alimentaria afecta directamente la salud humana en la vida diaria, ya que la preparación o el almacenamiento inseguros pueden provocar contaminación y enfermedades transmitidas por los alimentos. La encuesta de consumidores de PwC de 2025 re

La noticia

arXiv:2604.01444v3 Announce Type: replace Abstract: Food safety directly affects human health in daily life, as unsafe preparation or storage can lead to contamination and foodborne illness. PwC's 2025 consumer survey reports that 47% of respondents are comfortable using generative AI for meal planning, highlighting the need for safe food-related assistance. Yet the safety of large language models (LLMs) against malicious food-related requests remains underexplored. We introduce FoodGuardBench, an FDA-grounded benchmark with 2,124 seed malicious questions, 16,992 variants from eight jailbreak strategies, and 1,000 benign questions across five food-safety risk categories. Across twelve LLMs, the mean overall ASR is 25.51%, whereas the average per-model maximum across eight attacks is 60.59%, revealing