Designing AI agents to resist prompt injection

T1Covered by 1 source · first reported by OpenAI News at 11 Mar, 11:30 UTC

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