AgentMinder: intent-level runtime governance for AI agents

At VMware Explore 2026 on Aug 31, Broadcom launched AgentMinder and made it generally available. Its thesis is blunt: existing security models were designed for human users, but an autonomous AI agent is a "digital employee" that observes, decides, and acts on its own. What enterprises need is not better prompt-level guardrails but a traffic controller that stops overstepping at the action level.

The problem AgentMinder targets

Agents differ from ordinary software because they do not follow a rigid instruction set — they observe, think, decide, and act to reach a goal independently. That creates three risks: accidentally accessing restricted data, making unauthorized changes, and overstepping authority. Model guardrails only work at the content-generation layer; static permissions only answer "who this agent is." Neither answers the question that matters: "Has this agent been approved to do what it is about to do?"

AgentMinder's answer is to push authorization down to the action level: every agent becomes an enterprise-grade identity whose authority is bound to a declaration — its mission, permitted intents, approved tools, and authorized resources. Before an action touches an enterprise system, the agent must declare what it is trying to do, not just who it is.

Three capability pillars: identity, runtime, observability

Architecture choice: reuse the authorization stack, don't rebuild it

This is AgentMinder's most notable design decision. It is not another standalone "agent security platform": it deploys alongside existing LLMs — on-premises, in VPCs, or across public clouds — and integrates with existing authorization stacks through the AuthZEN standard. Enterprises can reuse the policy-enforcement endpoints they already run, without funneling traffic through a single SaaS chokepoint or rewriting identity infrastructure.

Broadcom positions it inside a broader agentic-AI security architecture alongside VMware vDefend (workload protection) and VMware Avi Load Balancer (threat detection), aimed at sensitive workflows in finance, HR, and IT. IDC's Securing the Agentic Enterprise perspective makes the same call: this shift demands unified governance, continuous authorization, and real-time telemetry so that every agent action is observable, attributable, and reversible.

The scale numbers: runtime governance is not a bottleneck

The usual objection to governance layers is "it won't scale." Broadcom answered with itself: AgentMinder carries roughly 36M customer-related and 7M workforce-related API calls per day across more than 20M customer identities and 72K workforce identities, with multi-region active-active architecture delivering zero downtime even during maintenance. CIO Alan Davidson credits it with providing "chain of custody" between developers and multiple agents and skills.

For decision-makers the takeaway is direct: the engineering hurdle for runtime governance has been crossed. It is not a research prototype but a commercial product running at tens of millions of calls per day. The remaining question is not "can it be done" but "who writes the intent declaration clearly."

What this means for enterprises

First, the unit of governance is moving from "permission" down to "intent." Provisioning a service account and static roles used to pass for governance; the product frontier is now action-level — every tool call must answer whether it fits the agent's declared mission. Second, you do not have to throw away your identity stack: interoperability standards like AuthZEN let a governance layer grow on top of existing authorization systems — a low-cost path for enterprises that have already invested in IAM and PAM. Third, auditability becomes a hard requirement: when agents act autonomously, regulators and auditors will want a full chain of custody — who, why, and what the agent did — not a model scorecard.

Three steps to act

References

• Broadcom: Broadcom Unveils AgentMinder, An Enterprise Solution for AI Agent Governance and Runtime Control (2026-08-31) — https://www.globenewswire.com/news-release/2026/08/31/3353342/19933/en/broadcom-unveils-agentminder-an-enterprise-solution-for-ai-agent-governance-and-runtime-control.html
• Broadcom: AgentMinder product page — https://www.broadcom.com/agentminder
• The Technology Express: Broadcom Launches AgentMinder for AI Governance (2026-09-01) — https://thetechnologyexpress.com/broadcom-launches-agentminder-for-ai-governance/