LATIDIA · Investigación

Erosión enmascarada: cómo las puertas de aceptación pierden salvaguardas redundantes en los agentes que se mejoran a sí mismos

arXiv:2610.38392v1 Anuncio Tipo: nuevo Resumen: Los agentes LLM de autoaprendizaje editan su propio arnés, las indicaciones y reglas en torno a un modelo fijo, y mantienen una edición solo si una puerta de aceptación la califica más alto en un lote de tareas

La noticia

arXiv:2610.38392v1 Announce Type: new Abstract: Self-improving LLM agents edit their own harness, the prompts and rules around a fixed model, and keep an edit only if an acceptance gate scores it higher on a batch of tasks. Such a gate sees what the harness does under the current model, not what it would still prevent if one safeguard failed. When safeguards overlap, deleting one is masked: the others still block the harm, so the gate decides on task success alone and redundancy erodes unseen. We measure harm for every subset of four written safeguards in two agent environments and run 411 improvement loops under ten gate rules, mostly as controlled