Traditional cybersecurity best practices, such as access management, secure protocols, penetration testing, vulnerability management, and intrusion detection, help protect AI systems and infrastructure. AI development, deployment, and operations necessitate a comprehensive security architecture that safeguards the infrastructure against traditional cybersecurity threats, including social engineering, intrusions, data breaches, malware, credential compromises, and privilege escalations. Therefore, the security architecture should include the concepts of zero trust, least privilege, and layered defenses. However, the shift towards AI-enabled applications increases complexity and opens up new avenues for cyberattacks. Therefore, organizations must integrate AI-specific security measures, including adversarial testing, model robustness techniques, data lineage tracking, and LLM firewalls.