Agent Grid is now in public preview, bringing identity-based Zero
Trust network security to AI agents on supported Azure environments. Security
teams can govern communications among agents, models, tools, and external
services through centrally managed policies. Enforcement operates outside the
agent’s trust boundary, without agent code changes or customer-managed
sidecars.
Preview capabilities include TLS inspection, fully qualified domain
name (FQDN) filtering, URL filtering, granular Model Context Protocol (MCP)
tool filtering, and Azure Content Safety integration for Prompt Shields.
Administrators can allow specific tools on an approved MCP server while
blocking sensitive operations, restrict outbound destinations, and inspect
supported content for prompt-injection attacks.
Agent Grid applies common policies across onboarded ACI Agent Groups,
AKS Agent Sandbox using Kata Containers, and native Kubernetes workloads on
Azure Kubernetes Service (AKS). Verified workload identities, mutually
authenticated encrypted connectivity, and centralized enforcement visibility
help security teams protect agent workflows across supported hosting
environments.
Learn more: