AI is a double-edged sword: Why you need new controls to manage risk

This article warns that AI introduces unique security risks beyond traditional controls, detailing threats to models (poisoning, evasion, theft), malicious inputs (prompt/code injection), and AI supply chain issues (code execution, malware, lateral movement); it urges building AI-specific visibility, standards (e.g., CoSAI, model signing), investing alongside AI initiatives, and hardening ML workflows by inventorying AI assets, expanding TI and vendor risk questions, and scanning models—especially risky serialized formats like Pickle—for hidden behaviors.