How Do You Defend Against AI That Can Hack?

AI agents represent a fundamental shift in cybersecurity, rendering traditional signature-based and behavioral detection tools obsolete because they target neither malicious code nor human actors. As enterprise software increasingly adopts agentic processes, the attack surface expands exponentially, leaving security teams struggling to define normal behavior. Current safety guardrails, while necessary to prevent misuse, often impede defenders during incident response by triggering false positives or blocking legitimate diagnostic queries. To navigate this, organizations are shifting toward flexible, multi-model strategies, including hosting open-weights models to maintain control. While AI introduces unprecedented risks, it simultaneously provides defenders with powerful new capabilities to automate threat research and infrastructure analysis. Ultimately, the industry is entering a period of rapid transformation where visibility into agentic processes and the ability to adapt to unpredictable software behavior are paramount for effective defense.