Securing AI to Benefit from AI

Securing AI to benefit from AI argues that artificial intelligence can greatly amplify cyber defence, but only if the systems underpinning it are secured with clear governance, strong identity controls, and visibility into decision-making. The piece highlights that every model, script, or autonomous agent operating in production represents a distinct identity that must be governed, authenticated, and auditable end to end.

It calls out the rise of agentic AI systems that can act without human intervention and stresses that trust must be bound to identity, policy, and monitoring across the entire lifecycle. It also lays out six control domains—Access Controls, Data Controls, Deployment Strategies, Inference Security, Monitoring, and Model Security—that map to recognised frameworks such as NIST AI Risk Management and OWASP Top 10 for LLMs, to help translate guidance into practice.

According to Frank Kim, SANS Institute Fellow, the discussion will be explored in more depth at SANS Surge 2026, with emphasis on keeping AI systems safe to depend on.