LATIDIA · Investigación

AgentPerturb: Evaluación de la robustez del agente de LLM bajo perturbaciones de interacción controladas

arXiv: 2610.21624v1Tipo de anuncio: nuevo Resumen: las tareas del mundo real a menudo son complejas y ruidosas, lo que plantea desafíos sustanciales para los agentes de LLM. Sin embargo, las evaluaciones existentes carecen de un marco sistemático para caracterizar conjuntamente

La noticia

arXiv:2610.21624v1 Announce Type: new Abstract: Real-world tasks are often complex and noisy, posing substantial challenges for LLM agents. Yet existing evaluations lack a systematic framework for jointly characterizing potential perturbation targets and types. We use this two-dimensional view to conduct a unified study that quantifies the robustness of current frontier models, traces the mechanisms behind perturbation-induced failures, and draws implications for robustness-oriented training. Through controlled evaluations on APEX-Agents and Toolathlon, we find that perturbations consistently degrade performance, while the dominant vulnerabilities vary across task environments. Trajectory-level analysis reveals that agents rarely recognize altered inputs and often propagate corrupted information without verification. A focused cost analysis further shows that successful perturbed