Decoding the world of cybersecurity

HYPR launches identity controls for AI agents

HYPR has launched AgentPass to link autonomous AI agents with accountable human supervisors. The platform uses identity policies, a controlled gateway and passkey approvals to govern actions taken through connected…

HYPR launches identity controls for AI agents
Summary
  • HYPR AgentPass assigns governed AI agents to human supervisors and evaluates their requests against identity and access policies.
  • A gateway can restrict connected tools, hold sensitive operations for approval and record the identities associated with agent activity.
  • The first generally available version was released in November, with protection depending on agents and tools being routed through the supported controls.

Enterprises can now link autonomous AI agents to named human supervisors and apply identity controls to their activity through a new governance platform from HYPR.

HYPR launched AgentPass on 18 November 2026, extending its identity assurance technology to software agents that use models and connected tools to perform tasks on behalf of employees.

The platform assigns governed agents to accountable people, evaluates requests before they reach connected services and can require additional authorisation for operations identified as sensitive.

HYPR's technical documentation confirms that the first generally available version was released in November. The offering follows an early access programme and supports governance arrangements for compatible agents operating on workstations and through cloud services.

Agent identities are connected to supervisors

Identity systems have traditionally distinguished employees from service accounts and other automated processes. AI agents introduce additional complexity because a single agent may interact with several applications while carrying out instructions originally issued by a person.

A coding agent, for example, may be authorised to inspect repositories, run development tools and request information from external services. Each operation may involve different permissions, even though the agent is carrying out one broader task.

Where the agent operates using persistent credentials, its effective access may continue independently of the employee who originally configured it.

AgentPass introduces a relationship between the agent, its supervisor and the identity provider used by the organisation.

The platform supports enterprise identity systems including Microsoft Entra, Okta, Ping and compatible OpenID Connect providers. Information about the supervisor and their roles can inform which tools the agent is permitted to use.

HYPR documents several mechanisms for establishing an agent's identity. An agent may use its own credential associated with an accountable owner, operate directly under a supervisor's identity or receive a scoped session token for a particular connection.

Under the scoped token arrangement, the gateway retains the underlying credential for the connected tool rather than exposing it to the agent.

That separation can reduce the need to distribute reusable service credentials directly to software performing automated tasks.

Gateway controls restrict connected operations

AgentPass operates as a control point between governed agents and the services they call.

The gateway evaluates requests against configured policies before forwarding them to supported destinations, including application programming interfaces, AI providers and Model Context Protocol tools.

Policies can restrict which tools an agent may use, prevent specified operations or hold a request until a supervisor authorises it.

The system also maintains records of the activity passing through the gateway, connecting requests with relevant identities, policies and decisions.

Such records can help establish which agent initiated an action and which supervisor was responsible for it when investigating unexpected access or changes to a business system.

However, gateway controls only apply to activity routed through the supported enforcement mechanisms. An agent that can reach an external service through an unmanaged connection may remain outside the system's observation and policy decisions.

Deployment therefore requires organisations to identify the agents and tools in use, establish how their traffic is routed and determine whether direct connections need to be restricted separately.

Permissions also require appropriate design. A technically valid policy can still authorise excessive access if the organisation grants an agent more capabilities than its intended tasks require.

Passkeys support decisions on sensitive actions

Where an action requires human authorisation, the gateway can suspend the request and send it to the agent's supervisor.

The supervisor can review the action through the HYPR mobile application or management console and approve or deny it before execution proceeds.

HYPR supports approvals signed using a passkey, linking the decision to the person providing authorisation.

Its documentation distinguishes between a production configuration requiring signed approvals and an optional mode in which approvals may be recorded without the same signature requirement.

Consequently, the presence of passkey support should not be confused with a guarantee that every installation requires a signed approval for every agent operation.

Policies determine which actions need intervention, while the deployment configuration controls how that approval is recorded.

Human approval also depends on context. A supervisor who cannot identify the affected application, requested operation or possible consequences may approve an action without understanding what the agent intends to do.

Organisations need to decide which operations warrant additional scrutiny and how those decisions should be presented to the people responsible.

Accountability continues after deployment

AgentPass allows authorised users to stop governed agents by preventing subsequent requests from proceeding through the gateway.

This creates an intervention mechanism when an agent behaves unexpectedly or its supervisor's permissions are withdrawn.

The ability to block future requests does not necessarily reverse actions already completed by an agent or terminate operations occurring outside the gateway's control.

Investigations may still require examination of the applications, records and credentials involved in earlier activity.

For enterprises deploying growing numbers of agents, maintaining a record of ownership can also support periodic access reviews and offboarding.

An agent created for an employee who later changes roles may need different permissions, while an abandoned automated process could retain access long after its original purpose has ended.

Identity governance can help address those conditions, although it cannot independently establish that an agent's business decisions or generated outputs are correct.

HYPR's generally available release provides mechanisms for attributing and controlling supported automated activity. The effectiveness of those controls depends on deployment coverage, policy design and the organisation's ability to monitor what its agents do after receiving authorised access.

×