How would you design a red-teaming strategy to identify vulnerabilities in a newly deployed generative AI model at TikTok?