Ory launches Agent Security to enforce identity inside AI agents

Ory said Wednesday it has made Ory Agent Security generally available, a control plane that enforces identity and access policies directly inside AI agent harnesses. The release is meant to close a growing security gap as enterprises deploy agents faster than they can govern them.

Why it matters: - AI agents are moving into enterprise systems faster than many organizations can govern them. - Ory Agent Security is designed to control what an agent is allowed to do at the moment an action is taken, not just monitor activity after the fact. - The platform creates an accountability trail from the human owner to the agent, sub-agents and tools.

What happened: - Ory announced the general availability of Ory Agent Security on Sept. 10, 2026. - The product is described as the first Agent IAM control plane built to secure AI agents at the point where they take action. - The platform enforces identity and access policies inside the agent harness before a tool action executes. - The system allows or denies actions, records the decision, and binds actions to an accountable owner and delegated agent identity. - Ory Agent Security is available now through the company's announcement and can run through Ory Network or as a self-managed deployment.

The details: - Ory Agent Security supports 11 agent harnesses and 14 agent SDK frameworks. - Supported frameworks include Claude Code, LangChain/LangGraph, OpenAI Agents SDK and Vercel AI SDK. - The platform is designed to work with custom-built agents as well as common commercial frameworks. - The product starts in observe mode so teams can see humans, agents, sub-agents and tools without blocking developers. - Organizations can then move to enforcement on their own timeline. - Policies can range from broad guardrails to fine-grained permissions. - Every decision is captured through OpenTelemetry for audit and security workflows. - The platform can be self-managed in cloud, on-premises or air-gapped environments. - Ory Agent Security is built on OAuth 2.1, OIDC, Zanzibar-style permissions and OpenTelemetry. - Ory said the standards-based approach is meant to fit into existing identity, security and infrastructure investments. - Ory’s identity infrastructure supports more than 3.25 billion identities, 4.4 billion transactions per day and more than 34,000 production deployments.

Between the lines: - Ory is positioning agent security as an identity problem first, not just a network or monitoring problem. - The emphasis on observe mode suggests Ory is trying to reduce adoption friction for teams that are not ready to block agent actions immediately. - Support for self-managed and air-gapped deployments signals demand from regulated and security-sensitive organizations. - TeleGlobal’s endorsement shows the pitch is aimed at managed services partners that want deterministic controls for enterprise AI rollouts. - John Poptsis, managing director at TeleGlobal, said Ory is a critical component of the TeleGlobal Compass framework because it adds enterprise-grade identity and access controls for agentic AI. - Poptsis also said the integration helps customers move from AI experimentation to secure, governed AI adoption at scale.

What's next: - Ory says organizations can adopt the platform in observe mode first and then turn on enforcement when they are ready. - The company is betting that enterprises will increasingly standardize agent identity and authorization across multiple frameworks. - Broader adoption will likely hinge on whether security teams accept identity-enforced controls inside agent execution paths.

The bottom line: - Ory is trying to make AI agents governable by putting identity and access control inside the agent itself, where action happens.