Publicación automática · Texto en el idioma de la fuente. How ChatGPT defends against prompt injection and social engineering by constraining risky actions and protecting sensitive data in agent workflows.