LATIDIA · Ciberseguridad

Rellenar previamente el canal de razonamiento: Ataques de prefijo de salida en los LLM de razonamiento

arXiv:2609.29775v1 Tipo de anuncio: nuevo Resumen: los modelos de lenguaje grandes (LLM) consumen y producen una sola secuencia de texto; por lo tanto, si se puede agregar texto al comienzo de la respuesta del LLM, es decir, un prefijo de salida, entonces

La noticia

arXiv:2609.29775v1 Announce Type: new Abstract: Large Language Models (LLMs) consume and produce a single sequence of text; hence, if text can be added to the beginning of the LLM's response, i.e., an output prefix, then all subsequent tokens will be conditioned on it. This output-prefix attack technique is a cheap black-box prompt injection. Prior work has shown this type of attack can reliably jailbreak non-reasoning models. Most reasoning models add an intermediate scratchpad reasoning step before the assistant's final response. The ability to edit this reasoning channel is exposed by some APIs and attack vectors can be leveraged for reasoning injection attacks. We present the first systematic, controlled study that