LATIDIA · Ciberseguridad
RAS: Medición de la seguridad de los LLM mediante la alineación de los rechazos
arXiv:2606.25750v2 Tipo de anuncio: sustitución Resumen: La evaluación repetida de la seguridad de los grandes modelos lingüísticos de código abierto requiere generar y valorar respuestas para numerosos puntos de control relacionados. Presentamos RAS (Alineación de rechazos)
WhatsApp ↗Telegram ↗
La noticia
arXiv:2606.25750v2 Announce Type: replace Abstract: Repeated safety evaluation of open-weight large language models requires generating and judging responses for many related checkpoints. We introduce RAS (Refusal Alignment Score), a representation-based, reference-relative indicator of harmful-request refusal computed by the five-stage SafeVec procedure. SafeVec extracts layer-wise refusal directions from an aligned reference model, measures their alignment in related targets, and calibrates a score from 0 to 100 using behavioral attack success rates (ASR) from reference and calibration models. Subsequent target scoring uses input forward passes alone. Across 18 held-out variants from the Llama, Gemma, and Qwen families, RAS correlates with $100(1-\mathrm{ASR})$ (Pearson 0.934, Spearman 0.851) and achieves 83.0% pairwise ordering agreement. Across