LATIDIA · Ciberseguridad

Puntos de referencia adversarios y ataques de BenchDoor: explotación de incentivos de referencia para ataques de puerta trasera

arXiv: 2610.27533v1Tipo de anuncio: nuevo Resumen: los puntos de referencia de aprendizaje automático actúan como guías para el desarrollo de modelos y estimulan la creación de conjuntos de datos de capacitación destinados a mejorar las puntuaciones de los puntos de referencia. Presentamos BenchDoor,

La noticia

arXiv:2610.27533v1 Announce Type: new Abstract: Machine learning benchmarks act as guideposts for model development and spur the creation of training datasets intended to improve benchmark scores. We introduce BenchDoor, a family of data poisoning attacks that exploits this incentive. Beyond preserving benign performance, BenchDoor requires benchmark gains that encourage adoption of data that also implants a backdoor. The adversary either manipulates a benchmark's training or test data, creating an adversarial benchmark, or supplies an independent training dataset while leaving the benchmark unchanged. We formalize five attack scenarios, including training on rewritten test examples and applying similarity-based decontamination. Across multiple-choice Q&A, text classification, and open-ended Q&A with Qwen and Gemma models,