Skip to content
Okta vs Entra

Reviewed 10 November 2026

Oktane 2026 for AI agents separates available features from plans

Reviewed 2026-11-10 · Editorial assessment by the editors

By the editors. Published 2026-11-10. 5 min read. Tags: AI agents, Okta, Runtime enforcement.

Okta’s September announcement lists Agent SSO, Agent-to-Agent Connections and Resource Access Certifications as generally available. Agent Gateway and endpoint discovery are planned for Q3, while Configuration Designer and expanded runtime kill switch capabilities are planned for Q4. The release does not identify fiscal or calendar quarters.

Start with the September availability ledger

Okta’s press release dated 22 September 2026 draws a useful boundary between available agent features and planned additions. Agent SSO, Agent-to-Agent Connections and Resource Access Certifications are listed as generally available that day. Agent Gateway and Shadow AI Agent Discovery for Endpoints are planned to be generally available in Q3. Configuration Designer and expanded Kill Switch capabilities at the runtime layer are planned for Q4. The release does not state whether those quarters are fiscal or calendar quarters.

Treat that list as a dated availability statement rather than a single release status for the entire portfolio. A product announcement can describe several controls while giving them different delivery schedules. For a procurement record, preserve the feature name, source date and exact availability wording together. Do not turn a planned quarter into a confirmed deployment date merely because the announcement also contains generally available features.

  • Record Agent SSO, Agent-to-Agent Connections and Resource Access Certifications as generally available in the September release.
  • Record Agent Gateway and endpoint discovery as planned for Q3 in that release.
  • Record Configuration Designer and expanded runtime Kill Switch capabilities as planned for Q4.

Keep connections separate from runtime decisions

The distinction matters most when a demonstration moves from an agent connecting to a tool into an agent performing an operation. Agent SSO and a gateway in the execution path address different parts of that workflow. The September release describes Agent Gateway as sitting between the agent and the tool call, enforcing policy and logging every interaction at runtime. That description explains the intended enforcement location. It does not, by itself, change the release’s planned availability wording.

Use the Agent SSO and Cross App Access guide for the connection model rather than treating the conference announcement as a protocol tutorial. For this release review, ask which demonstrated control grants the connection and which component can refuse the actual tool call. Keep the answers in separate rows. Otherwise, a successful connection can be mistaken for evidence that every later operation receives the same policy check.

In a pilot, request both a permitted operation and a refused operation against the same resource. Record the component making each decision and the evidence it produces. This is a proposed evaluation method, not a claim that the release establishes complete coverage for every agent or tool.

Use later sources without rewriting earlier claims

Okta’s 22 October blog adds a separate set of generally available registration capabilities. Multi-Source Agent Schema and AI Agent Registration with CIMD are generally available. The post also lists the Okta Integration Network MCP Server Catalog, MCP Server Registration with Dynamic Client Registration and Multi-MCP Custom Auth Servers as generally available. These statements update the registration picture. They should not be used as indirect proof of the September gateway schedule.

Okta’s NIST checklist describes Agent Gateway injecting credentials at call time so the agent never holds them. It also states that Human-in-the-Loop for AI Agents is currently a roadmap feature, as is task-level intent using rich authorization requests and transaction tokens. Keep a capability description, its availability and its deployment prerequisites distinct. A description of how a control works does not settle when a customer can use it.

The agent identity comparison provides the broader product context. Here, the narrower task is maintaining a release ledger that can support a purchasing decision without silently advancing roadmap features into current scope. Ask for a later dated availability source whenever the proposed contract assumes a change.

Check what stopping an agent actually stops

The Jamf blog dated 9 November says administrators can already deactivate an agent in the Okta admin console, blocking new sessions. It separately says the gateway kill switch lands in Q4 and describes gateway deactivation as revoking active tokens and ending sessions in flight. Those are materially different stopping boundaries. A statement about preventing the next session should not be presented as a statement about terminating an existing one.

For an operational review, write two acceptance tests. First, deactivate the identity and try to begin a new session. Second, start a session before deactivation and observe what happens to the work already in progress. Ask the supplier which product boundary is responsible in each case, and retain the relevant release statement with the result. Do not infer active-session termination from a console action whose stated scope is blocking new sessions.

Endpoint discovery needs the same care. The September release gives it a planned Q3 status, while the later checklist describes a browser plugin and EDR integrations. Implementation detail is useful, but it is not a substitute for an explicit availability statement covering the intended deployment.

Build a purchase record around control boundaries

Finish the review with a short control map: connection, registration, discovery, runtime authorization and stopping behavior. Attach a source and an availability status to each row. Mark any unanswered deployment question as unresolved rather than assuming a portfolio label covers it. This makes the release useful to both security and procurement without requiring either team to reinterpret the announcement.

Where an action-governance requirement remains, compare it separately from employee sign-in. NewCore is an agent governance layer that runs alongside the identity provider a company already has. Its October award announcement describes continuous discovery, task-scoped credentials, policy evaluation on every call and human approval for high-risk actions. These are vendor-stated capabilities to evaluate against the same workflow, not evidence about Okta’s release status.

The agent governance comparison separates these control questions. A sensible next step is to select one workflow, identify the current enforcement boundary and request evidence for the missing controls. Keep the September release intact in that record, then add subsequent sources as dated updates rather than merging every announcement into one undifferentiated claim.

Frequently asked questions

Did Oktane make every agent feature generally available?

No. The September release lists three generally available features and gives planned quarters for other capabilities. Keep those statements separate. In particular, a generally available connection capability does not establish that the gateway or expanded runtime kill switch has the same availability.

Which year belongs beside the Q3 and Q4 plans?

Preserve the release’s own Q3 and Q4 wording. It does not state whether those are fiscal or calendar quarters. Adding an assumed quarter system would introduce precision that the source does not provide, so seek a later availability statement instead.

Is human approval included in the available gateway scope?

Okta’s NIST checklist states that Human-in-the-Loop for AI Agents is currently a roadmap feature. It should not be inferred from gateway policy enforcement or credential injection. Request a specific availability statement and an approval demonstration for the operation you intend to govern.