Why AI Agents Break the GenAI Security Model with Devvret Rishi - #770
AI agents that take actions break the static-guardrails-plus-human-approval security model enterprises rely on.
“why human-in-the-loop review can become security theater when agents operate at scale”
Rubrik's Dev Rishi argues that agentic AI's ability to plan, call tools, and act at machine speed invalidates traditional pre-written guardrails and one-at-a-time human approvals. Enterprises instead need runtime enforcement, agent observability, policy-aware governance, and recovery mechanisms like agent rewind. This reframes AI security as a runtime, defense-in-depth problem rather than a static-rules one.