How to Red Team a Gen AI Model.

This article details red teaming to address generative artificial intelligence risk. The article explains red teaming, who within an organization should be part of the effort, and its degradation objectives. Degradation objectives include bias in the models and privacy harms. Next, the article explores manual and automated attack testing including code injection and content exhaustion. Lastly, the coordination of these efforts, with an emphasis on documentation and legal privilege, are discussed.