Descope Launches Agentic Identity Hub 3.0 to Secure Shadow AI Agents

Descope Inc. released Agentic Identity Hub 3.0 today, a new version of its platform designed to secure artificial intelligence agents operating within corporate environments. The company states that this update specifically addresses the growing problem of shadow AI, where employees build or adopt agents faster than security teams can inventory them, often using shared credentials without recorded ownership.

According to SiliconANGLE, the release focuses on bringing these unauthorized or unmanaged agents under centralized control. Descope cites Verizon Communications Inc.’s 2026 Data Breach Investigations Report, which found that 45 percent of employees use AI on corporate devices. The report indicates that approximately two-thirds of those users sign in to AI services using personal accounts, creating significant visibility gaps for information technology teams.

Addressing Shadow AI Visibility

To gain control over these agents, Descope introduced an Agent Onboarding Portal that requires employees to request access through a self-service catalog. Each request must specify the data sources and Model Context Protocol servers the agent will connect to. No agent is provisioned until a reviewer approves the request, ensuring every approved agent has an accountable owner and scoped access.

Descope argues that this workflow is sufficiently fast to prevent teams from bypassing it. The platform addresses the challenge that many agents are built by third parties, such as Anthropic PBC’s Claude or Anysphere Inc.’s Cursor, which organizations cannot directly instrument. By integrating natively with MCP gateways, including the Agentic AI Foundation’s open-source agentgateway project, Descope places a single identity and policy layer in front of all agents and servers.

Authentication and Policy Enforcement

The update also targets the security risks associated with static client secrets, which often leak into code repositories and are rarely rotated. Under new client authentication options, an agent signs a short-lived assertion with a private key that remains on the client side. This approach eliminates the need for long-lived static credentials that are vulnerable to exposure.

Internal agents receive access similar to single sign-on. Users authenticate once through their existing corporate identity provider, and subsequent access to each MCP server is granted without additional login screens or consent prompts. Descope stated that policy evaluation occurs on every request, ensuring continuous compliance even for external clients like OpenAI Group PBC’s ChatGPT.

What should you do?

Security teams should audit current AI agent usage to identify shadow IT risks. Organizations need to establish clear ownership for any automated agents interacting with corporate data. IT departments should evaluate their Model Context Protocol gateway configurations to ensure policy enforcement is possible.

What to do and how to stay safe: Descope

• Inventory all AI agents currently in use by employees to identify those operating without recorded ownership or proper credentials.
• Implement a formal request and approval process for new AI agents before they are connected to corporate data sources or MCP servers.
• Replace static client secrets in agent configurations with short-lived authentication methods to reduce the risk of credential leakage in code repositories.
• Configure identity providers to enforce policy evaluation on every request made by AI agents to ensure continuous access control.

Step-by-step guide: Prompt Injection Warning Signs: How to Spot AI Manipulation

General security guidance from the Threat Times newsroom. It is not confirmed advice from the organisations named in this story.

Frequently asked questions

What is the main focus of Descope Agentic Identity Hub 3.0?

The update focuses on securing shadow AI agents built or adopted by employees, ensuring they have accountable owners and scoped access through centralized policy enforcement.

How does the platform handle third-party AI agents?

It integrates natively with MCP gateways, placing a single identity and policy layer in front of agents and servers from external providers like OpenAI or Anthropic.

Does this version change how agents authenticate?

Yes, it introduces client authentication options where agents sign short-lived assertions with private keys, replacing static client secrets that are prone to leakage.