LATIDIA · Ciberseguridad

LLM de peso abierto de Jailbreaking a través de perturbaciones de incrustación aleatorias

arXiv: 2610.07125v1Tipo de anuncio: nuevo Resumen: Si bien los modelos de peso abierto han disfrutado de un progreso constante en las capacidades y una amplia adopción en múltiples dominios, su seguridad sigue siendo una preocupación importante. Una característica clave

La noticia

arXiv:2610.07125v1 Announce Type: new Abstract: While open-weight models have enjoyed steady progress in capabilities and wide adoption across multiple domains, their safety remains an important concern. One key feature is the ability to refuse or deflect harmful, malicious, or insensitive prompts. In this paper, we expose safety vulnerabilities across six common open-weight LLMs of various sizes that consistently lead to harmful or unsafe responses on the JailbreakBench benchmark dataset. Our proposed attack, Perturbed Embedding Vector (PEV), is a simple and fast "jailbreaking" technique that is cheaper than prior approaches, which typically require gradient computations, per-prompt optimizations, or altering internal weights of the models. PEV just adds independent Gaussian noise in