The breakthroughs and improvements that we uncover produce new means of thinking, new connections, and new industries. The second goal of confidential AI will be to develop defenses towards vulnerabilities which might be inherent in the use of ML products, for instance leakage of private information by means of inference queries, or creation of ad