Agent Identity governance: the enterprise register your agent fleet is missing · Enterprise Agentic AI Insights
Google Cloud shipped Agent Identity, Agent Gateway and Model Armor. 96% of enterprises run agents in production, only 12% can govern them. The six-column fix before EU and California transparency law lands Aug 2, 2026.
You can open your HR system and name every person on payroll. Try naming every AI agent running inside your cloud accounts right now, what it touched last, and who approved it. Google Cloud consolidated Vertex AI into the Gemini Enterprise Agent Platform this year and shipped exactly the tooling that question needs: Agent Identity gives every agent a unique cryptographic ID, enforced below the application tier and mapped to an authorization policy, so every action an agent takes is traceable. Agent Gateway acts as a control plane for agent-to-agent and agent-to-tool traffic, blocking any outbound call to a host not registered in Agent Registry, the platform's centralized catalog. Model Armor extends that perimeter with runtime protection against prompt injection and sensitive data leakage. This is not a policy document. It is infrastructure, shipped and live. The gap it is meant to close is bigger than most executives assume. A 2026 OutSystems survey of 1,900 IT leaders found 96% of enterprises already run AI agents in production. Only 12% say they can govern them. Separate identity-security research puts numbers on why: non-human identities, including AI agents, now outnumber human identities roughly 45 to 1 inside the average enterprise. Ninety-two percent of organizations lack full visibility into those identities. Mean monitoring coverage across production agents sits at 52%, meaning close to half of all agents running in enterprises today operate with no monitoring at all. Seventy-eight percent have no documented policy for creating or retiring an AI identity, and 84% could not currently pass a compliance audit focused on agent behavior or access. This is not a Google-specific story. AWS shipped Bedrock Guardrails and a FinOps Agent for cost and risk visibility earlier this summer. Snowflake shipped Cortex AI Gateway to control agent sprawl and spend at the data layer. Three separate ecosystems have now shipped real governance infrastructure inside a matter of months. The tooling is not the bottleneck. Adoption of the tooling is, and it is falling behind at the same rate agent counts are rising: the most active enterprise deployers are already running 76 to 100 agents, and that number is doubling every quarter. Here is the board-level turn. EU AI Act Article 50 transparency obligations bind August 2, 2026, three days from this post, and apply instantly to every in-scope system regardless of when it was deployed. On the identical date, California's AI Transparency Act also becomes operative. Any enterprise serving both EU and California users faces two independent transparency regimes activating on the same day, and neither regulator will accept "we don't have a full list of our agents" as an answer. The fix is not a platform migration. It is a one-page register built this week from data your platforms already track: every agent in production, the cryptographic identity or credential it runs under, the gateway or guardrail policy attached to it, the systems and data it can reach, a named owner, and a yes-or-no flag for EU or California-facing use. Six columns turn a scattered agent fleet into an audit-ready inventory, and they turn "we run agents" into "we can account for every agent we run," the sentence your CISO, your board, and two regulators all want to hear by Monday. The move: stand up the Agent Identity Register this week, tie it to the identity and gateway data your platform (AWS, Google, or Snowflake) already tracks, and flag EU/California-facing agents before August 2. Sources: Google Cloud Blog, Introducing Gemini Enterprise Agent Platform; Infosecurity Magazine, Google Introduces Unique AI Agent Identities; Google Cloud Documentation, Agent Gateway overview and Model Armor integration; GCP Study Hub, Vertex AI Is Now Gemini Enterprise Agent Platform; Security Boulevard, The Agent Identity Problem; Akeyless, 2026 AI Agent Identity Security Survey; Cloud Security Alliance, The AI Agent Governance Gap; VentureBeat, Snowflake launches Cortex AI Gateway; Sidley, EU AI Act Article 50 compliance guidance; Glacis, US State AI Laws Tracker 2026.