LATIDIA · Ciberseguridad
Los agentes de IA siempre pueden caer en las trampas de las inyecciones de «prompts»
arXiv:2605.17634v2 Tipo de anuncio: sustitución Resumen: La inyección de «prompts» es la vulnerabilidad más crítica en los agentes de IA implementados. A pesar de los avances recientes, demostramos que el paradigma de defensa predominante (separación de datos e instrucciones)
WhatsApp ↗Telegram ↗
La noticia
arXiv:2605.17634v2 Announce Type: replace Abstract: Prompt injection is the most critical vulnerability in deployed AI agents. Despite recent progress, we show that the prevailing defense paradigm (data-instruction separation) both fails to detect attacks that operate through contextual manipulation and degrades contextually appropriate behavior. We then recast prompt injection via the lens of Contextual Integrity (CI), a privacy theory that judges information flow compliance with contextual norms. This explains types of attacks that current defenses attempt to patch and predict advanced ones future agents will face. We develop unique benign and attack scenarios that force an agent to violate the norms by (1) misrepresenting the flow, (2) manipulating norms, or (3)