Broadcom has introduced AgentMinder, a generally available platform designed to apply identity, authorization and audit controls to enterprise AI agents as they access data, invoke tools and complete workflows.
The company said AgentMinder treats an agent as an enterprise identity and binds its authority to a declared mission, permitted intents, approved tools and authorized resources. Its cloud-native gateway authenticates tokens and evaluates each tool call against a policy engine before directing traffic to approved backends.
The product targets a governance gap created when AI agents move beyond producing text and begin taking actions in business systems. Broadcom said the platform uses OpenTelemetry-based observability to record agent sessions and actions, including evidence intended to support audit and anomaly detection.
AgentMinder can be deployed with models running on premises, in virtual private clouds or in public clouds. Broadcom said it integrates with existing authorization infrastructure through the AuthZEN standard, allowing organizations to retain policy enforcement endpoints rather than route traffic through a single SaaS service.
For IT and security teams, the operational question is whether agent permissions can be constrained and reviewed with the same rigor as human and service identities. Broadcom positions the product for higher-risk uses in finance, HR and IT, but organizations will still need to define their own policies, permitted tools and incident-response processes.
Broadcom said AgentMinder supports VMware vSphere Kubernetes Service, Google Cloud Platform and other cloud-native Kubernetes environments. It is generally available now.