Securing Agentic AI: New Challenges for Businesses

Security teams face new challenges as businesses use more agentic AI. These autonomous systems operate across identities, data, and infrastructure at machine speed. The issue is traditional security controls were not designed for this. When problems occur, it's difficult to detect and contain them. This often happens because organizations deploy agentic AI without fully establishing guardrails, observability, and governance. Diana Kelley, CISO of Noma Security, notes that excessive autonomy and unintended data exposure are becoming common. This isn't due to malicious intent, but because AI operates in environments not designed for autonomous decisions. Kelley emphasizes that many companies are still in the proof-of-concept phase with agentic deployments. They want to ensure proper behavior before full production. This situation highlights the urgent need to secure AI systems before they are widely implemented.