Broadcom Launches AgentMinder to Control Enterprise AI Agents

Broadcom has launched AgentMinder, an enterprise security and governance platform designed to control how autonomous AI agents access tools, applications and data.

The company announced AgentMinder on August 31, 2026, at VMware Explore 2026 in Las Vegas. Broadcom said the product is generally available immediately.

AgentMinder is designed around the idea that enterprise AI agents need an identity, defined permissions and controls that remain active while they perform tasks. Instead of relying only on static access permissions, the system evaluates an agent’s identity, intended action and surrounding context before allowing a request to proceed.

Broadcom describes this as a way to control what an agent is authorized to do rather than simply determining whether the underlying user or application has access.

The platform includes a cloud-native AI gateway that sits between agents and the resources they are attempting to use. Broadcom said the gateway authenticates tokens and routes requests only to authorized backends.

Its policy engine can evaluate an agent’s identity, the user’s identity, the intended action, the requested tool and the resource being accessed. The company said the gateway can enforce policies at runtime rather than depending entirely on permissions configured when an agent was created.

According to Broadcom, AgentMinder also treats an agent’s declared mission and intent as part of the authorization process. This is intended to prevent an agent with legitimate credentials from using those credentials for activities outside its approved purpose.

The platform also provides observability and audit capabilities through OpenTelemetry. Broadcom said this gives security, risk and platform teams visibility into agent sessions and individual actions, including audit trails, anomaly detection, operational telemetry and policy decisions.

Broadcom said AgentMinder can integrate with existing authorization infrastructure through the AuthZEN authorization standard, allowing organizations to use existing policy enforcement endpoints rather than replacing their authorization systems.

The product is designed to support large language models running on-premises, in virtual private clouds and in public clouds. Broadcom also said it can operate across cloud-native Kubernetes environments, including VMware vSphere Kubernetes Service and Google Cloud Platform.

Broadcom said it is already using AgentMinder internally for its own agentic pipeline. The company reported nearly 36 million customer-related API calls and about 7 million workforce-related API calls per day, alongside more than 20 million customer identities and 72,000 workforce identities.

Broadcom said the internal deployment uses a multi-region, active-active architecture and has maintained availability during maintenance and upgrades. These figures are Broadcom’s reported deployment statistics.

The launch is part of a wider set of AI-security announcements from Broadcom around VMware Cloud Foundation and its VMware Private AI Cloud strategy.

Broadcom also announced new agentic AI security capabilities for VMware vDefend and VMware Avi Load Balancer. The company said vDefend can discover MCP servers, large language models, datastores and tools while helping identify unauthorized “shadow AI” deployments.

Broadcom said vDefend can also generate intrusion prevention and detection signatures through an agentic AI pipeline and continuously monitor traffic across VMware Cloud Foundation.

For Avi Load Balancer, Broadcom said the new capabilities are intended to restrict agents from unauthorized MCP tools, inspect agent transactions and block malicious activity such as remote code execution and file injection. The company also said the technology can detect anomalous agent traffic and help prevent sensitive information from being exfiltrated.

Broadcom is also expanding agent capabilities within the VMware Tanzu Platform. The company describes Tanzu as the agent platform for VMware Private AI Cloud.

Broadcom said the Tanzu agent foundation uses a deny-by-default runtime in which agents do not receive access to APIs, networks, MCP servers or the internet unless that access is explicitly granted. The platform also includes an isolated credential store designed to keep credentials away from agents themselves.

AgentMinder is intended to work with this wider architecture by providing identity, authorization, runtime policy enforcement and observability for agent activity.

Broadcom has also connected the technology to its FinOps strategy. The company’s CloudHealth team said it is integrating AgentMinder’s OpenTelemetry data to associate AI spending with users, projects, agents, missions and business cases.

CloudHealth also described capabilities involving policy-based routing, budget fallbacks and circuit breakers for AI spending. These should be treated as capabilities described by Broadcom rather than evidence that every feature is currently generally available within AgentMinder.

Broadcom said the rapid adoption of agentic systems is increasing the number of connections between AI agents, models, APIs, databases and other enterprise services. It specifically identified protocols such as the Model Context Protocol (MCP) and agent-to-agent communication as part of the expanding environment.

In this environment, an AI system can potentially retrieve information, call an internal API, interact with an application or pass work to another agent. Broadcom’s approach is to place authorization and policy controls around those actions while maintaining records of what the agents do.

The company’s strategy also aligns with recent industry research on securing agentic AI. Broadcom cited IDC research published in June 2026, titled Securing the Agentic Enterprise: Part 2 — Assessing the Impact of Agents on Key Security Domains, which examines the security implications of autonomous agents and the need for identity-driven governance.

AgentMinder is not presented as an AI model or an agent-building platform. Broadcom positions it as an enterprise control layer covering agent identity, authorization, runtime enforcement and observability.

The company has not published a specific public price for AgentMinder in the launch material. Broadcom directs organizations seeking more information about the product to contact the company.

For enterprises deploying autonomous agents, the core proposition behind AgentMinder is that authorization should continue while an agent is performing a task, rather than ending when credentials are issued. Broadcom is extending that approach across its private-cloud, network and application-security products as it builds a broader security architecture for agentic AI.