Microsoft's red team uncovers AI vulnerabilities by thinking like adversaries to strengthen generative AI systems.