Skip to main content
Version: 11.5.0

Require a Person for High-Value Agent Actions

Who this is for: the Operator or Admin who writes your organization's policies, with an Admin for the gateway settings near the end.

Purpose​

This playbook gives the running order for letting agents make routine calls on their own, while HYPR AgentPass™ holds named sensitive actions. A held action waits until a person approves it with a passkey.

Each task is a summary. The detail, with screenshots, field values and caveats, is on the page the step links to.

When you finish:

  • The actions you chose are held until a supervisor approves them.
  • Each approval is bound to that person and that action.
  • The policy set stays readable as it grows.

Architecture​

A policy evaluates each request as it passes through the gateway. A Require approval match holds the request, and asks the supervisor on their phone. For the diagram and the request path, see Architecture: The Request Path.

Prerequisites​

  • Role: Operator or Admin. Drafting, checking and promoting the rule need Policies: Add and Policies: Edit, which Operator and Supervisor hold for their own policies, and Admin for every policy. Require Signed Approvals needs Settings: Edit, which only Admin holds. Attaching evaluators needs Session watch: Manage, which only Admin holds.
  • At least one connector carrying agent traffic
  • Supervisors with an approval passkey
  • A drafting model, if you want to draft policies with AI

For the full list, see Prerequisites.

Procedure​

Complete these tasks in order.

Find What Is Ungoverned​

Open Policies → Coverage. The table shows which connector and tool pairs a rule reaches, and which no rule reaches. Pick the sensitive actions from the ungoverned rows.

See Policy Coverage.

Draft the Rule in Observe Mode​

On Policies, click New policy. Choose Start from scratch to write the rule yourself, or Describe it in your own words to draft it from a plain-English goal. Set the action to Require approval, and leave the enforcement mode on Observe. In Observe mode, the rule records what it would have held, without holding anything.

See Creating a Policy, Authoring a Policy with AI and Setting Up AI Authoring.

Check the Impact and the Overlap​

Open the rule's page from Details on its row. Read its plain-language summary on the Definition tab, and what it matched on the Decisions tab. Then open Policies → Redundancy, and confirm no existing rule already covers the same traffic or sits ahead of yours.

See Understanding Your Policies and Redundant and Conflicting Rules.

Promote the Rule to Enforce Mode​

On Policies, click the rule's State badge, and change it from Observing to Enforcing. From the next matching request, the action is held until its supervisor answers.

See How Policies Are Evaluated and Observe First, Then Enforce.

Require Signed Approvals​

This task needs an Admin. Open Settings → Gateway settings → Passkeys (FIDO2). Set the passkey approval policy, and set Require signed approvals to Required (production). Each approval is then bound to the supervisor's passkey, and to the action approved.

See Connecting to Your HYPR Tenant.

Keep the Policy Set Healthy​

Review Policies → Suggested for rules the gateway proposes from observed traffic. Ask an Admin to attach evaluator agents to sessions, so suspicious behavior is surfaced without blocking.

See Suggested Policies and Session Watch.

Validate​

Have an agent attempt one of the chosen actions, and check the following:

  • The request is held, and the supervisor's phone shows the approval.
  • Approve releases the action, and Deny stops it.
  • The rule's Matches count on the All policies tab of Policies increases, and the request appears on the rule's Decisions tab as Sent for approval.

See Approving and Denying Agent Actions.

Logs and Audit Trail​

Every held request and its outcome is on the console's Approvals page. The request itself, with the policy that held it, is in the Request Log, and the rule's decision is on Activity → Policy decisions.

See Approvals in the Console, Request Log and Policy Decisions.