GPT-Red: internal automated red team for detecting prompt injections

Developers introduced an internal, automated red team tool called GPT-Red, designed to uncover models' vulnerabilities to prompt injections at scale. The tool helps build stronger defenses before wider deployment, reducing security risks.