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.

Source: OpenAI — Published March 11, 2026 at 11:30 UTC — Category: Models