LATIDIA · Ciberseguridad

Con los ojos vendados: Cómo mantener en secreto la moneda de auditoría ante los agentes de LLM

arXiv:2610.37055v1 Anuncio Tipo: nuevo Resumen: La certificación basada en auditoría limita la tasa de violación de un agente LLM mediante la auditoría de una fracción aleatoria $\varepsilon$ de acciones y el reponderado de las etiquetas auditadas; su validez

La noticia

arXiv:2610.37055v1 Announce Type: new Abstract: Audit-based certification bounds the violation rate of an LLM agent by auditing a random fraction $\varepsilon$ of actions and reweighting the audited labels; its validity is meant to rest only on the defender's audit coin. We show that it also rests on a confidentiality property: the coin must not interfere with the agent's view before the action commits, at a time the agent chooses. We call this pre-commit noninterference (PNI) and measure its violation by $D_{\mathrm{A}}$, the total-variation distance between the agent's views under the two coin values. PNI is necessary and sufficient for an unbiased certificate against adaptive agents, the worst-case bias is $(1-\varepsilon)D_{\mathrm{A}}$,