Oktane 2026 brought identity and security leaders to Caesars Forum in Las Vegas from September 22 to 24. Okta’s flagship conference centered on securing AI agents, governing privileged access, and turning customer identity into a growth engine.
Across the keynotes, Okta argued that AI agents are entering production faster than many companies can identify, authorize, monitor, or deactivate them.
Key Announcements
A Blueprint for the Secure Agentic Enterprise
Okta first published its blueprint for the secure agentic enterprise in March, organized around where an organization’s agents are, what they can do, and what they are doing. At Oktane, the new Blueprint Alliance expanded it into an open, multi-vendor reference architecture and added a fourth question about how security teams should respond.
Founding members include AWS, CrowdStrike, Databricks, Google Cloud, Salesforce, and ServiceNow. In his keynote, Okta CEO and co-founder Todd McKinnon said agents need “deep, broad, persistent access” to systems to deliver value, and that the challenge is granting it securely at scale.
Agent SSO Gives Agents First-Class Identities
Agent SSO, generally available since August 24 and included in core Okta SSO plans at no extra cost, brings the Cross App Access standard into Okta’s main product. Supported agents are registered in Universal Directory, governed through established administrative workflows, and issued short-lived tokens for each connection.
Many agents still connect through static API keys, custom integrations, and isolated OAuth grants. Agent SSO gives administrators a familiar control model while keeping the employee, the agent acting for that employee, and the receiving application distinct. Anthropic’s David Soria Parra, a co-creator of the Model Context Protocol, joined McKinnon on stage to make the case for the standard, and Claude now supports Cross App Access with Okta as its first identity provider.
Runtime Controls Move Into the Path
Okta also expanded Okta for AI Agents, generally available since April 30, with Shadow AI Agent Discovery for Endpoints, a visual Configuration Designer for agent connections, and support for third-party identity providers. Much of the runtime layer is still on the roadmap, with Agent Gateway and endpoint discovery slated for general availability in the third quarter and the Configuration Designer and a Kill Switch that revokes an agent’s active tokens at the gateway following in the fourth.
Agent Gateway sits between agents and the tools or Model Context Protocol servers they call. This in-path position lets organizations evaluate the agent’s identity, the person behind the request, applicable policies, and permitted actions before issuing a short-lived credential.
Governance, Privilege, and Threat Detection Converge
Okta also highlighted additions across Okta Identity Governance, Okta Privileged Access, and the identity threat detection it gained by acquiring Permiso Security earlier this year. Governance adds Advanced Entitlement Management for AWS and Intelligent Request Recommendations, planned for early access by the fourth quarter, while Privileged Access extends zero standing privilege to workloads and AI agents, with general availability due by the first quarter of 2027.
The product direction ties access approval to ongoing behavior, so authorization keeps adjusting after the first grant. Ely Kahn, Okta’s chief product officer, described the target as dynamic, real-time control across people, AI agents, and non-human workloads, and Okta tied the push to Verizon data attributing 39% of breaches to credential misuse.
Auth0 Pushes Identity Toward Growth
Auth0 announcements widened the conference story beyond workforce security. The Identity Conversion Suite, entering early access in late September, lets brands recognize visitors before they sign up and A/B test sign-up flows, and Auth for UCP, part of AI Identity for Commerce, lets shoppers complete purchases through agents inside Google Gemini using the Universal Commerce Protocol.
For B2B software companies, B2B Connect layers enterprise federation onto an existing identity system with no user, app, or token migration, and the already available Tenancy-as-a-Service handles customer onboarding and tenant management. Auth0 Chief Product Officer Gareth Davies framed the opportunity around protecting transactions while reducing the identity work that slows product teams.
Strategic Insights
Agent Inventory Becomes a Board-Level Dependency
Companies cannot govern software actors they cannot locate, identify, or connect to accountable owners.
Harish Peri, Okta’s senior vice president and general manager for AI security, demonstrated how Okta can find an unregistered agent and MCP server on an employee’s endpoint and register both in Universal Directory. Agents now arrive from several directions, whether an employee attaches an assistant to a business application, a developer builds one inside a project, or a vendor embeds one in a purchased product.
Shadow AI Agent Discovery for Endpoints reflects how the discovery problem has moved beyond cloud consoles and approved development platforms. Matt Ellard, Okta’s senior vice president and general manager for EMEA, said shared integrations could exchange threat signals and federate agent registries, helping companies find deployments outside approved channels.
Authorization Needs Human and Machine Context
A useful authorization trail must show the agent, its owner, the user represented in a specific interaction, the requested resource, and the policy applied. Without that chain, security teams may see a technically valid token while missing whether the underlying action had legitimate business authority.
Cross App Access and Agent SSO address the connection layer by moving authorization decisions toward the enterprise identity provider. Agent Gateway extends that model into execution by checking policy each time an agent invokes a tool.
Runtime Enforcement Draws the New Boundary
Agents can act continuously, call multiple systems, and hand work to other agents, which makes periodic reviews too slow for higher-risk processes. Once the roadmap ships, gateway enforcement, token revocation, and a deterministic Kill Switch will give security teams several intervention points during an active session.
Organizations will need to establish which agents, applications, and Model Context Protocol connections can sit behind enforceable controls, then test whether revocation actually stops downstream activity.
Identity Spending Faces a Consolidation Test
The conference tied governance, privileged access, and identity threat response more closely together.
That combination could reduce separate policy stores and disconnected investigations, but buyers should demand proof across their own environments. Directory records, cloud entitlements, privileged sessions, agent activity, and threat signals must retain enough context to support a defensible decision.
Okta’s positioning also puts pressure on adjacent vendors, since identity providers, endpoint platforms, API gateways, and AI governance tools increasingly touch the same agent transaction. For buyers, integration quality becomes the deciding factor.
The Undercurrent
Across the announcements, Okta worked to move identity deeper into execution, building a chain from discovery and connection through runtime authorization to response that reaches far beyond conventional sign-on.
Okta repeatedly said no single vendor can secure agents alone, yet its product announcements placed Universal Directory, Agent Gateway, governance, and threat response at the center of the operating model. That leaves an open blueprint resting on vendor-specific enforcement.
Start With the Agent Inventory
An agent inventory comes first, recording each agent’s owner, purpose, data access, and expiration date. From there, teams can replace standing credentials where possible, restrict agents to approved tools, and carry a common session identifier across identity, delegation, and tool-invocation events.
Procurement reviews should ask which features are generally available today, which agent types are covered, whether third-party identities can be governed, how quickly active access can be revoked, and what evidence reaches audit and incident-response systems.
What’s Next
Roadmaps should separate foundational controls from newer runtime functions. Identity registration, accountable ownership, short-lived credentials, and tested deactivation procedures belong in near-term implementation plans.
Agent gateways and cross-vendor registries deserve controlled pilots tied to production workflows and explicit success criteria. Each pilot should show that when an agent exceeds its authority, the organization can identify the actor, reconstruct the authorization chain, stop the activity, and prove that access stayed revoked.
