LATIDIA · Ciberseguridad

Jev como barandilla de seguridad de LLM: un estudio empírico de filtrado de entrada, revisión selectiva de salida y rechazo excesivo

arXiv:2611.08845v1 Tipo de anuncio: nuevo Resumen: Los filtros de seguridad externos pueden evitar que las solicitudes inseguras lleguen a un modelo de lenguaje grande (LLM), pero también pueden rechazar consultas legítimas. Presentamos una e empírica de extremo a extremo

La noticia

arXiv:2611.08845v1 Announce Type: new Abstract: External safety filters can prevent unsafe requests from reaching a large language model (LLM), but may also reject legitimate queries. We present an end-to-end empirical evaluation of Jev, a probabilistic decision model, as a safety guardrail for a Qwen2.5-1.5B-based pipeline. Three configurations are compared on 970 benchmark prompts: an unfiltered LLM, Jev input filtering, and tiered Jev input filtering with selective output review. Across 720 prompts labeled harmful (520 AdvBench and 200 XSTest-unsafe), the fraction recorded as answered decreases from 527/720 (73.2%) without filtering to 17/720 (2.4%) with input filtering. On 250 XSTest-safe prompts, however, non-answer/refusal increases from 3/250 (1.2%) to 27/250 (10.8%). Tiered filtering