Red teaming is a cybersecurity and AI safety practice where teams simulate attacks or adversarial scenarios to identify vulnerabilities and weaknesses in AI systems. In AI contexts, it involves testing models for harmful outputs, bias, or security flaws.

NYD Application: We red team our AI implementations to ensure they don't produce harmful recommendations, leak sensitive data, or exhibit unexpected behaviors in client environments.

Example: "Our red team testing revealed that the code generation model could be prompted to suggest insecure authentication patterns, so we added safety filters."