Redefining Enterprise Cybersecurity in the Age of Adversarial AI (M01a)
This talk addresses the critical issue of adversarial AI and its potential to exploit AI systems, leading to significant financial and reputational risks for organizations. A two-pronged approach is proposed to protect the integrity of AI model behavior: rigorous penetration testing of AI models to identify vulnerabilities and targeted monitoring of model inputs to detect and mitigate potentially harmful manipulations. The talk examines the vulnerability of both traditional prediction-based models and newer generative AI models to adversarial attacks, with an emphasis on real-world examples and enterprise case studies.