openai.com6 months agoDesigning AI agents to resist prompt injectionHow ChatGPT defends against prompt injection and social engineering by constraining risky actions and protecting sensitive data in agent workflows.Visit openai.com1BookmarkAdd to collection