The most common causes of AI-related breaches include familiar threats: exploited software vulnerabilities (21%), AI model flaws (19%), insider threats (18%), and misconfigured cloud settings (16%).