LATIDIA · Ciberseguridad

Lo que sobrevive a la autoevolución: marcas de procedencia en arneses de agente copiados

arXiv: 2610.21458v1Tipo de anuncio: nuevo Resumen: arneses de agente, el indicador y el código de herramienta en torno a un modelo base, fuga. Las 512.000 líneas de Claude Code se reflejaron horas después de un error de empaquetado. Un plagiario puede pasar un le

La noticia

arXiv:2610.21458v1 Announce Type: new Abstract: Agent harnesses, the prompt and tool code around a base model, leak. Claude Code's 512,000 lines were mirrored within hours of a packaging error. A plagiarist can pass a leaked harness through a published self-evolver and ship the result as its own. We define this as a covert rewriting attack. No one targets a provenance mark, yet a large language model (LLM) proposer rewrites the file that carries it and a score keeps the better rewrites. We give the first measurement of marks under this attack, with three evolvers, five proposers and two environments. Published marks fail. KGW in the prompt survives GEPA's rewrites in