Newest papers

Agent Mosaic: Recognizing Confidentiality Leaks in Agentic LLM Systems

Daniel Ruiz, Luca Luceri, Emilio Ferrara · 2026 · arXiv (Cornell University)

DOI: 10.48550/arxiv.2610.30418 · cited by 0 works (OpenAlex) · open-access PDF

Spanish

Este efecto mosaico crea un riesgo crítico para los agentes que operan sobre datos confidenciales. Cuantificamos su realización con 30 ejemplos de datos sintéticos de seguridad cibernética, médicos y militares, produciendo 10,800 respuestas finales de tres agentes LLM en simulación controlada y 1,440 respuestas de agentes en vivo en dos aprovechamientos de código abierto. Incluso bajo consultas benignas diseñadas para ser respondibles de un solo documento, GPT-5.6-Sol, el modelo de mejor rendimiento probado, revela compuestos prohibidos en el 7,08% de las respuestas (95% CI: 3,75-10.67%) en simulación, con peores tasas de divulgación en harnesses en vivo. Añadiendo cabecillas de clasificación visibles y reglas de agregación explícita reduce sustancialmente las tas

Machine translation (Meta M2M100); the English original follows.

Abstract

Modern LLM agents can retrieve information from accessible documents, and then disclose a sensitive conclusion assembled from them. This mosaic effect creates a critical risk for agents operating over confidential data. We quantify its realization with 30 synthetic cybersecurity, medical, and military examples, yielding 10,800 final answers from three LLM agents in controlled simulation and 1,440 answers from live agents in two open-source harnesses. Even under benign queries designed to be answerable from one document, GPT-5.6-Sol, the best performing model tested, discloses forbidden composites in 7.08% of answers (95% CI: 3.75-10.67%) in simulation, with worse disclosure rates in live harnesses. Adding visible classification headers and explicit aggregation rules reduces pooled disclosure rates substantially, yet unsafe disclosures persist in 32.37% of simulated answers (95% CI: 28.33-36.19%) across all intents despite these safeguards. Furthermore, increased refusals do not guarantee confidentiality, since significant refusals reveal the prohibited conclusion regardless. These results point to a troubling conclusion: sensitive data leaks via mosaic-style events are an observed confidentiality failure mode in the evaluated settings at non-negligible rates. They motivate stronger controls on what agents may infer and disclose in sensitive data environments with mixed-clearance users.

Read in: English · Spanish · Chinese · Hindi · Arabic · Portuguese · French · German · Japanese · Russian · Korean · Indonesian · Vietnamese

Topics: Confidentiality, Computer security, Internet privacy, Point (geometry), Computer science, Realization (probability)

Sources: OpenAlex (CC0), arXiv, Crossref. Agent Exchange