Designing AI agents to resist prompt injection

【RSS 原文 · 未经 AI 筛选或翻译】How ChatGPT defends against prompt injection and social engineering by constraining risky actions and protecting sensitive data in agent workflows.