What are the primary security vulnerabilities unique to AI systems, and how do you mitigate them for The Boston Consulting Group engagements?