Broadcom AgentMinder: Identity-First Governance for Agentic AI — Architecture, Enforcement, and Why It Matters Now

Your AI agents are making decisions, calling APIs, and delegating work to other agents — all without a single identity check. That gap between what agents can do and what they should do is exactly what Broadcom built AgentMinder to close.

Announced at VMware Explore 2026 on August 31 and now generally available, AgentMinder↗ is Broadcom's answer to a question most enterprises are only starting to ask: how do you govern software agents that act autonomously?

This article breaks down why agentic AI governance matters right now, how AgentMinder's architecture works under the hood, and what it means for organizations already running — or planning to run — autonomous agents at scale.

The Problem Nobody Planned For

Enterprise AI has moved fast. In less than two years, we went from chatbots answering help-desk tickets to multi-agent systems that book travel, approve expenses, triage security alerts, and write production code — with minimal human oversight.

The security models we rely on weren't designed for this. Traditional identity and access management (IAM) assumes a human is behind every action. OAuth tokens, RBAC, SAML assertions — they all trace back to a person who logged in. But when an AI agent calls an API, who is the principal? When that agent spawns a sub-agent that calls a different API, what governs the chain of trust?

Most organizations today handle this with one of two approaches, and both are inadequate. Some embed static API keys in agent configurations — the agent becomes a service account with broad permissions and zero context about why it's making a given call. Others rely on the human user's session token, which means the agent inherits the full scope of human permissions regardless of the task at hand.

Neither approach answers the fundamental questions: What is this agent trying to accomplish? Is it allowed to do this specific thing right now? And can we prove it later?

What AgentMinder Actually Is

AgentMinder is a governance platform purpose-built for agentic AI. It gives every agent a verifiable identity, constrains what each agent can do through mission-bound policies, and produces compliance-grade audit trails for every action taken.

Think of it as IAM rebuilt from scratch for a world where the majority of API calls come from software agents, not humans.

Broadcom didn't build this in a vacuum. The platform draws on decades of identity infrastructure — the same team behind the Identity Security Platform (IDSP), Privileged Access Management (PAM), and Layer7 API Gateway. AgentMinder extends that foundation into the agentic domain, which is why it ships with native integration points for all three.

The architecture is organized around three pillars.

Pillar 1: Identity and Intent

Every agent registered with AgentMinder receives a cryptographically verifiable identity — not just a name or an API key, but a proper credential that can be validated, rotated, and revoked. This is the "who" part.

But identity alone isn't enough. AgentMinder introduces a concept called mission objects — structured declarations of what an agent is supposed to accomplish and what boundaries it must operate within. A mission object might say: "This agent is authorized to query the CRM and draft a customer response, but it cannot access financial records or send messages without human approval."

Mission objects travel with the agent. When Agent A delegates a task to Agent B, the child agent inherits the parent's mission constraints. It can operate within those boundaries — but never exceed them. This is enforced cryptographically, not by convention.

This solves one of the hardest problems in multi-agent architectures: delegation without privilege escalation. Every link in the chain has a bounded scope, and that scope can be audited end-to-end.

Pillar 2: Runtime Enforcement

Identity and intent declarations are only useful if something enforces them at the point of action. That's where AgentMinder's AI Gateway comes in.

The AI Gateway is a protocol-aware enforcement point that sits in the request path between agents and the tools or backends they call. For every action an agent attempts, the gateway evaluates four dimensions:

• Who — the verified identity of the requesting agent
• What tool — the specific API, function, or resource being called
• Which resource — the data or system being accessed
• Why — the stated intent from the agent's mission object

Policy decisions happen in real time. The gateway authenticates the agent's token, validates it against the mission constraints, checks the action against the configured policy, and either permits or blocks the call — all before the request reaches the backend.

The policy decision point (PDP) is pluggable. Organizations with existing authorization stacks built on AuthZEN↗ can integrate their own PDP rather than ripping and replacing what they already have. This is a meaningful design choice — it means AgentMinder fits into existing governance infrastructure rather than demanding that enterprises start over.

Pillar 3: Observability and Audit

Every action that passes through AgentMinder — permitted or denied — generates a structured audit event. The observability layer is built on OpenTelemetry↗, which means it integrates natively with whatever monitoring stack an organization already runs (Splunk, Datadog, Grafana, Elastic, and so on).

The audit trail captures the full context: which agent acted, what mission it was operating under, which tool it called, what resource it accessed, and the policy decision that was made. For multi-agent workflows, you get the complete delegation chain — from the originating human request through every agent handoff to the final action.

This is what turns AgentMinder from a security tool into a compliance tool. Regulated industries — finance, healthcare, government — need to prove not just what happened, but why it was allowed to happen. A flat log of API calls doesn't cut it. A full-context audit trail with cryptographically verifiable identity chains does.

Under the Hood: Architecture Decisions That Matter

A few architectural choices stand out and are worth understanding if you're evaluating this for production deployment.

Cloud-native, deploy-anywhere. AgentMinder runs on-premises, in a VPC, or on public cloud. There's no mandatory SaaS chokepoint. For organizations with data sovereignty requirements or air-gapped environments, this is a hard requirement that many competing approaches can't meet.

Protocol-aware, not just proxy-aware. The AI Gateway doesn't just route HTTP traffic. It understands the protocols that agentic frameworks use — including emerging standards like MCP (Model Context Protocol) and A2A (Agent-to-Agent). This means it can make policy decisions at the semantic level (what tool is being invoked, with what parameters) rather than just at the network level (which URL was called).

Action-based, not role-based. Traditional RBAC says "this identity has the Admin role, therefore it can do everything an Admin can do." AgentMinder's model is more granular: policies evaluate the specific action being attempted, in the context of the agent's current mission, against the specific resource being accessed. Two agents with identical roles can get different policy decisions based on their mission constraints.

Separation of concerns. Policy authoring, policy evaluation, and enforcement are architecturally separate. Security teams author policies. The PDP evaluates them. The gateway enforces them. This separation means you can update policies without redeploying infrastructure, and you can swap the PDP without changing the gateway or the agents.

Broadcom's Own Scale Numbers

Broadcom doesn't just sell this — they run it internally. The numbers from their announcement give a sense of the scale the platform handles:

• 36 million customer-facing API calls per day
• 7 million workforce API calls per day
• 20 million+ customer identities under management
• 72,000 workforce identities

These aren't lab benchmarks. They're production numbers from one of the world's largest technology companies running AgentMinder against its own agent workloads.

As Clayton Donley, VP and GM of Broadcom's Identity Security division, put it at the launch: this is about extending proven identity infrastructure to cover the agents that are already operating inside enterprises — whether those enterprises have governance policies for them or not.

Why This Matters Now

The timing of AgentMinder's release isn't accidental. We're at an inflection point where enterprises are moving from AI experimentation to production agent deployment. Gartner, Forrester, and every major analyst firm are flagging agentic governance as a critical gap. The risks are real:

• Data exfiltration — an agent with broad API access can leak sensitive data through tool calls, even without malicious intent
• Privilege escalation — multi-agent delegation chains can accumulate permissions that no single agent was intended to have
• Compliance gaps — when regulators ask "who authorized this action," pointing at a cron job that triggered an LLM chain isn't an acceptable answer

The organizations that get ahead of this will be the ones that treat agent identity and governance as first-class infrastructure — not as an afterthought bolted on after the first incident.

Getting Started

AgentMinder 4.1 is generally available now. Broadcom offers it as part of its identity security portfolio, and it integrates natively with existing Broadcom infrastructure — IDSP for identity lifecycle, PAM for privileged access, and Layer7 for API management.

For organizations already invested in the Broadcom ecosystem, the integration path is straightforward. For those coming in fresh, AgentMinder's pluggable architecture and standards-based design (OpenTelemetry, AuthZEN, OAuth 2.0) mean it can work alongside existing identity providers and authorization frameworks.

The product documentation is extensive — over 2,000 pages covering deployment, configuration, policy authoring, and API reference. If you're evaluating, start with the product brief↗ and the Agentic Identity Fabric white paper for the conceptual foundation, then move to the deployment guides for your specific topology.

---

Bottom line: The agentic AI wave is here, and the security models we've been using for twenty years weren't built for it. AgentMinder is Broadcom's bet that identity-first governance — verifiable agent identities, mission-bound policies, and compliance-grade audit trails — is the right foundation. Given the scale they're already running it at internally, it's a bet worth paying attention to.