Google Locks Agentic OAuth Classification

The News

Google Workspace developer release notes say that starting November 11, 2026, Workspace MCP servers reject requests from apps whose OAuth client is not marked as agentic. OAuth clients not marked by November 12 remain non-agentic and cannot call Workspace MCP servers. The affected surface is enterprise web infrastructure for Workspace integrations. The mechanism is agentic OAuth gating across Gmail, Drive, Chat, Sheets, and Slides MCP server access.

The OPTYX Analysis

This is a governance and web infrastructure signal because Google is turning agent identity from a descriptive label into an access-control boundary. The systemic mechanism is credential-purpose separation, where agent workflows must use OAuth clients explicitly declared for AI-powered agents rather than reusing general application clients. Strategically, this creates a cleaner enforcement layer for consent, auditability, abuse response, and future policy differentiation between human-operated apps and delegated agents. The change matters because enterprise adoption of agents now depends on identity architecture, not only model capability or user permissions.

Enterprise Impact

The exposed operator is the Workspace administrator, identity architect, integration owner, or SaaS vendor maintaining MCP-connected workflows. The vulnerability is silent integration failure when existing OAuth clients serve both agent and non-agent features, or when production deployments missed the classification deadline. The required control is an OAuth client inventory that separates agentic and non-agentic use, maps each client to scopes and MCP surfaces, and updates deployment references. Next move is testing Gmail, Drive, Chat, Sheets, and Slides agent calls against new credentials before user-facing automations break.