LATIDIA · Ciberseguridad

Marca de agua conductual vinculada a la sesión para agentes LLM autónomos

arXiv:2610.38579v1 Anuncio Tipo: nuevo Resumen: Los agentes de modelos de lenguaje grandes (LLM) actúan cada vez más a través de interfaces web y herramientas externas. Producen trayectorias. Esas trayectorias pueden pasar por varias

La noticia

arXiv:2610.38579v1 Announce Type: new Abstract: Large Language Model (LLM) agents are increasingly acting through Web interfaces and external tools. They produce trajectories. Those trajectories may pass through several intermediaries. They can store, modify, or dispute those trajectories. In Web-based agent deployments, trajectories may pass through APIs, marketplaces, and tool servers. As such, a copied or disputed log may not by itself reveal which agent owner produced it. Behavioral watermarking can embed ownership evidence in agent action choices. However, existing methods mainly address detection, removal, rewriting, or imitation. They do not directly test whether repeated observation of watermarked trajectories enables cross-session forgery by exploiting learned action preferences. We formulate this learned-forgery