新闻 · OpenAI News

Designing AI agents to resist prompt injection

How ChatGPT defends against prompt injection and social engineering by constraining risky actions and protecting sensitive data in agent workflows.