CSO Online: Zero trust security programs lack controls for AI agent autonomous execution and judgment

According to CSO Online, zero trust architectures designed for human users are insufficient for agentic AI operating at scale. HashiCorp's Stephen Wilson describes AI agents as "really smart kindergartners" that know how to execute tasks but lack judgment about why they should, creating gaps in existing zero trust models; static credentials and standing privileges are inadequate for rapidly authorizing, limiting, and revoking agent permissions sometimes multiple times within a single workflow.