Why Are Large AI Models Being Red Teamed?

Large AI models are being red teamed to identify vulnerabilities and safeguard against potential harms. Red teaming is a military tool that is now being applied to AI in the public interest. It involves playing the role of the adversary to find vulnerabilities in AI models and fix them. AI red teaming is now part of AI public policy, and guidelines are being developed to support its deployment. However, red teaming is not a standalone solution, and other evaluation and mitigation techniques are needed to ensure the safety of AI models.