The second target of confidential AI will be to acquire defenses towards vulnerabilities that are inherent in the use of ML styles, for instance leakage of private information via inference queries, or development of adversarial illustrations. By enabling protected AI deployments in the cloud devoi