Solutions  ·  2026-06-29

Google Cloud VPC Service Controls — Agent Identity, MCP-Attribute Perimeter Rules, and Gemini Agent Platform Integration

SolutionsHigh impactGlobal
On June 27, 2026, Google Cloud announced three new VPC Service Controls capabilities targeting agentic AI workloads: (1) agent identities (SPIFFE-based, cryptographically attested) can now be added directly to ingress/egress perimeter rules as first-class IAM principals — individual or fleet-wide, with instant revocation if compromised; (2) conditional perimeter rules based on MCP attributes (mcp.toolName, mcp.method, mcp.tool.isReadOnly) enabling tool-level policy enforcement (e.g., allow read but block email-send); (3) native integration with Gemini Enterprise Agent Platform that auto-blocks all public internet access when Agent Platform is inside a VPC-SC perimeter.
VPC Service Controls is Google Cloud's established data-exfiltration prevention layer used by hundreds of enterprises. Extending it to treat AI agents as network-perimeter subjects — with MCP-level tool granularity — is a significant architectural shift that directly mitigates prompt-injection exfiltration (OWASP ASI01) and over-permissioned agent risks without requiring new products.
Google Cloud platform and security teams deploying agentic AI workloads should implement agent-identity perimeter rules now; especially urgent for workloads with access to Cloud Storage, BigQuery, or Workspace data.
Sources
Google Cloud BlogGoogle Cloud Agent Identity DocsGravityDevOps Analysis
See this in the live feed Explore related AI security and governance findings — updated every morning.
Open the feed →