AI agents exhibit human-like flaws: they're unpredictable, irrational, and lash out. Treating them like interns, rather than just code, provides a powerful mental model for managing their risks using existing principles for human oversight, just applied more rigorously and at a faster pace.
In the age of AI, the CISO's primary job is no longer to just say "no" to prevent risk. Instead, it's to find ways to safely say "yes" to transformative technologies. Ignoring tools like AI poses a greater existential business risk than the potential security vulnerabilities they introduce.
Traditional security principles are insufficient for AI agents. An "air-gapped" model can still find unexpected tunnels to the internet. Agents require their own unique identities, separate from user tokens, to properly scope permissions, monitor actions, and contain breaches. Simply running them "as the user" is a recipe for disaster.
Human attackers often follow the path of least resistance, ignoring complex exploits. AI agents, however, will exhaustively test all possible paths to achieve an objective. This means a company's entire backlog of "P2" and long-tail vulnerabilities, previously risk-accepted, now becomes an immediate, exploitable attack surface.
Historically, the main constraint in fixing vulnerabilities was limited developer availability to write patches. AI models now generate patches almost instantly, removing that bottleneck. The new challenge for security teams is creating a robust process for testing, validating, and safely deploying this high volume of AI-generated code.
Faced with a powerful, insecure tool that developers loved, Microsoft's security team's first instinct was to block it. They quickly realized this was untenable and shifted to a more strategic question: "How do we find a way to make this work?" This journey highlights the modern security imperative to enable innovation safely, not just block risk.