The useful question is what changes for users, developers or buyers, and whether the announcement stays industry context or becomes something people can actually use.

Okta has launched identity governance capabilities for AI agents, treating them as first-class identities within its platform. The new features aim to address the rapid but often ungoverned adoption of AI agents in enterprises by providing tools to discover, register, and assign ownership to these agents. This initiative responds to concerns about shadow AI and uncontrolled agent access, which can introduce security and compliance risks. The announcement underscores Okta’s focus on extending its identity security fabric to non-human entities.

The governance framework requires organizations to answer three core questions: identifying all agents in their environment, registering them in a centralized directory, and assigning human owners to close visibility gaps. Okta emphasizes controlling the connections agents rely on, such as MCP servers, APIs, and other agents, while enforcing access policies and credential vaulting to prevent lateral movement. The platform also introduces automated governance workflows and a kill switch to revoke access for rogue agents, supported by a full audit trail.

Key capabilities now generally available include continuous discovery of unknown agents, centralized registration of agents and MCP servers regardless of their origin, and the assignment of human owners to strengthen accountability. Okta also enables the provision of temporary, short-lived credentials to reduce reliance on long-lived tokens, alongside least-privilege access enforcement to protect critical systems. The kill switch feature allows organizations to deactivate agents and prevent new token requests if an agent behaves unexpectedly.

The announcement highlights early adopters such as Anthropic, Zoom, and Slack, and introduces Okta’s open XAA protocol to standardize secure, identity-governed agent connections. Okta also announced an integration with Amazon Bedrock AgentCore to provide an identity foundation for production-ready AI agents. The company provides resources, including a checklist and strategic blueprint, to help organizations govern AI agents and navigate the challenges of an AI-first enterprise.