What are the challenges in securing artificial intelligence systems from model extraction attacks?

One measure organizations can take to protect AI systems from model extraction attacks is to use techniques like regularization and noise injection to make it harder for attackers to reverse-engineer the underlying model. Additionally, implementing secure and robust encryption methods for storing and transmitting sensitive model information can add an extra layer of protection. Employing access controls, monitoring system logs for suspicious activities, and ensuring the physical security of the AI systems can also help prevent model extraction attacks. Regularly updating software and firmware along with conducting security audits and penetration testing can further enhance the security posture of AI systems against such threats.