Broadcom Launches AgentMinder to Give AI Agents Identities and Runtime Rules

Broadcom used VMware Explore 2026 in Las Vegas to launch AgentMinder, a product that treats every AI agent as an enterprise identity and checks each of its actions against a declared purpose before the action reaches a company system. The launch on August 31 is one of the clearest signs yet that agent governance is moving out of policy documents and into the infrastructure layer, where security and platform teams can enforce it. For CIOs and CISOs trying to get agents past security review, it puts one of the largest infrastructure vendors squarely into the agent control business.

What is Broadcom AgentMinder?

AgentMinder is generally available now, according to Broadcom's announcement. It comes from Broadcom's Identity Management Security Division, not its VMware cloud unit, and Broadcom describes it in three parts:

• Identity and intent. Each agent is registered as an identity with a declared mission, permitted intents, approved tools and authorized resources.
• Runtime enforcement. A cloud-native AI gateway sits in the path of every tool call. It authenticates tokens, applies a dynamic policy engine and routes traffic only to backends the agent is allowed to reach.
• Observability and audit. An OpenTelemetry-based layer records agent sessions and actions, keeps audit trails and flags anomalies.

The product deploys on VMware vSphere Kubernetes Service, Google Cloud and other standard Kubernetes platforms, and can run beside models on premises, in virtual private clouds or in public clouds. It plugs into existing authorization systems through AuthZEN, an emerging standard for connecting applications to external policy engines. Broadcom did not disclose pricing.

Clayton Donley, who runs the identity division, said in the release that AI has moved "from the pilot stage" into daily business operations.

Why does agent identity matter to enterprises?

Most enterprise access controls were built for people. A person who opens the wrong file usually notices and stops. An agent does not, and many early deployments give agents a shared service account with broad rights because it is the fastest way to get a demo working.

Speakers at AI Connect events have described that gap repeatedly. Ajit Gupta, CEO and founder of Midships, told The AI Enterprise Conference 2026 that "Traditional identity and access management says what you can access," while nothing checks whether an agent should be accessing something at that moment. That is close to the problem AgentMinder targets with its intent checks. In a AI Connect webinar on human-in-the-loop design and risk tiers, SAP's Shubham Trivedi called for least-privilege permissions per agent and said "the policy layer should sit outside the agent's own reasoning," so a bad action is stopped even when the model is convinced it is right.

A gateway that intercepts each tool call is one way to put policy outside the model.

How does it fit with the rest of VMware Explore?

AgentMinder arrived alongside VMware AI Factory, the software foundation of Broadcom's private AI offering. That product includes its own AI gateway for prompt routing, token rate limits and application authorization, plus isolated sandboxes for code that agents generate, with controls over which tools an agent can invoke, according to StorageNewsletter's report on the announcement.

The two products come from different Broadcom divisions, and the company did not explain in detail how they work together. Buyers evaluating both should ask whether policies are defined once or twice, and which gateway enforces what.

Has it been tested at scale?

Broadcom's main evidence is its own use. The company said AgentMinder supports more than 20 million customer identities and 72,000 workforce identities internally, with peaks of nearly 36 million customer-related and 7 million workforce-related API calls per day. Broadcom CIO Alan Davidson credited the platform with letting the company roll out its agentic AI pipeline company-wide.

Those are vendor figures from a single, friendly deployment, and they speak to throughput and uptime more than to how well intent-based policies catch an agent that goes off course, which is the harder test. Broadcom named no outside customers.

What should enterprise leaders do now?

Agent governance is shifting from "write a policy" to "enforce it at runtime," and vendors from identity, security, data and infrastructure markets are all claiming the job. Hemakumar Velugula of Xnode argued at The AI Enterprise Conference 2026 that the stack needs a control plane between applications and models, because "The access controls have to be reinvented. It has to go to the data layer."

Practical steps, regardless of vendor:

• Inventory agents as identities. Know which agents exist, who owns each one and what each is meant to do.
• Retire shared service accounts. Give each agent scoped, revocable credentials.
• Put enforcement in the path. Policy that lives only in prompts is advice, not control.
• Demand audit you can replay. Security and audit teams will want to show which agent did what, for whom, on a given day.

The next thing to watch is whether outside customers adopt AgentMinder beyond VMware shops, and whether standards like AuthZEN let enterprises keep one policy model as they mix agent platforms from several vendors.

Frequently asked questions

What is Broadcom AgentMinder?

AgentMinder is a generally available Broadcom product that manages AI agents as enterprise identities, enforces policy on each tool call through an AI gateway and records agent activity for audit.

Where does AgentMinder run?

Broadcom says it deploys on VMware vSphere Kubernetes Service, Google Cloud and other standard Kubernetes platforms, alongside models on premises or in private or public clouds.

Why do AI agents need their own identities?

Traditional access controls were designed for people and often give agents broad shared accounts. Per-agent identities let companies limit what each agent can do and trace its actions.

Filed under: Agentic AI, AI Governance & Trust, Security & Fraud