Descope adds approval and access controls for workplace AI agents

Descope has unveiled Agentic Identity Hub 3.0, a product update aimed at helping companies keep track of workplace AI agents and control what they can access. The practical shift is from agents appearing in company workflows unannounced to requiring an owner, review and defined permissions.

Watch Desk analysis

What happened

SiliconANGLE reports that employees can request agents through a new onboarding portal, listing the data sources and Model Context Protocol (MCP) servers they need. Nothing is provisioned until a reviewer approves it. Descope says approved agents receive an accountable owner and scoped access.

The update also adds integrations with MCP gateways, support for short-lived signed authentication assertions, and single sign-on for internal agents. Descope says access policies are checked on every request, and administrators can revoke an agent’s access centrally without reissuing credentials. Those are the company’s stated capabilities, not a report of independently tested results. SiliconANGLE’s report gives the product details.

What changes

• An approval step before provisioning
  Employees request agents through a catalogue, and a reviewer must sign off before access is granted.
• Clearer connections and permissions
  Requests can specify individual data sources and MCP servers, while approved agents get scoped access and a named owner.
• A control point for MCP agents
  Native integration with MCP gateways is intended to put one identity and policy layer in front of agents and MCP servers.
• Short-lived authentication
  Agents can sign a temporary assertion using a private key that stays on the client, rather than relying on long-lived client secrets.
• Central revocation and per-request checks
  Administrators can withdraw access in one place, while policies are evaluated on each request.

Why it matters

Companies cannot govern agents they do not know they have. A request-and-approval process, named owners and limited permissions could give security teams a more useful starting point than discovering an agent after it has been connected to company data. Short-lived credentials and central revocation also address a familiar security headache: access that persists long after its original purpose has ended.

The catch is that controls only help if teams use them and the system covers the agents and services they actually run. Descope’s announcement describes the intended machinery, not evidence yet that it closes the shadow-agent gap in practice.

Our read

This is a concrete attempt to apply identity management to agents, rather than asking security teams to trust a new class of software and hope for the best. The most useful details are the approval gate, scoped access and ability to revoke credentials centrally. Companies considering it should check which agents and MCP connections it can cover, and whether its review process is quick enough that staff will not simply route around it. Security controls are only charming when they work before the workaround does.

What to watch

• Which identity providers, agents and MCP gateways the product supports in practice.
• Whether organisations can inventory agents built or adopted outside approved channels.
• Evidence that approval and revocation work reliably across connected services.

Discussion spark: Should workplace AI agents need a named owner and human approval before receiving access to company data, or would that slow teams into using untracked tools instead?

Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.