LATIDIA · Ciberseguridad
Ocho de cada diez LLM filtran secretos de las indicaciones del sistema: Nevan, un estudio de «captura la bandera» con agentes
arXiv:2611.21956v1 Tipo de anuncio: nuevo Resumen: Auditamos la aplicación de la jerarquía de instrucciones en 363 modelos de entrada/salida de texto del catálogo de OpenRouter. La pregunta de investigación a la que respondemos con este experimento es qué modelo
WhatsApp ↗Telegram ↗
La noticia
arXiv:2611.21956v1 Announce Type: new Abstract: We audit instruction-hierarchy enforcement across 363 text-in/text-out models from the OpenRouter catalog. The research question we answer with this experiment is which models disclose a protected secret embedded in the system prompt, and through which techniques. Each model guards a secret 16-character flag set in its system prompt, while an artificial intelligence (AI) agent (OpenCode, powered by DeepSeek V4.1 Flash), steered by a human researcher, tries to persuade it to emit the flag. The flag was exfiltrated from 296 of the 363 models, plus four models released after the initial catalog freeze, for 300 captures in total. Of these, 170 involved a non-English language pivot,