HYPR AgentPass

Bind every AI agent to an accountable human.

Give your people the freedom to build with AI agents, without giving up control.

Five requirements for trusting AI agents

One control plane, from laptop to tool call.

Local agentsAgents on people's machines, routed through the gateway. Sync them from connected providers.
+ New workload agent
Agent
Acts as
Scopes
Active
A
Loan Intake Assistantsk-demo-…a41f
daniel.okafor@highlandsbank.comDaniel Okafor
inferencetools
A
Claude Code (Maya Thornbury)sk-demo-…5be8
maya.thornbury@highlandsbank.comMaya Thornbury
inferencetools
O
Codex (Tomás Rivera)sk-demo-…e17a
tomas.rivera@highlandsbank.comTomás Rivera
inferencetools
W
Nightly Loan Reconciliation
Role: AgentPassOperator1 connector
Acts as Maya ThornburySupervisor signed in through Okta

AgentPass · Local agents · Try it: pick a tab

How should this policy decide?The expression is what the gateway runs. Write it yourself, or describe what to stop and let AI write it — either way you end up editing the same expression.
Expression

Written by the HYPR built-in author · reference: none · attempt 1. Edit it freely — it is yours now.✓ Valid expression

Require approval — hold it until a supervisor approves⌄

Chosen by the HYPR built-in author from your description. Saved in OBSERVE mode so you can check the matches before enforcing it.

AgentPass · New policy · Try it: press Describe it instead

AgentFraud Triage AgentSupervisor: Priya Raman
AgentPass gatewayCheckingRequire approvalApprovedDenied
PolicyClosing a P1 ticket needs a person
PR
Checking the requestBefore anything reaches JiraWaiting for Priya RamanThe agent's supervisorApproved by Priya RamanSigned with her passkeyDenied by Priya RamanNothing was sent to Jira
ToolJiraNothing sent yetNothing sent yetTicket closedNothing sent
PRSupervisorPriya Raman
Wednesday, September 3010:21
AgentPass · nowApproval neededFraud Triage Agent wants to close a P1 incident ticket
10:21
Approval request
Fraud Triage AgentAn agent you supervise
wants toClose a P1 incident ticketToolJira
Face IDApprove with your passkey
ApprovedSigned with your passkey
Fraud Triage AgentClose a P1 incident ticket
DeniedThe request was refused
Fraud Triage AgentClose a P1 incident ticket

Try it: allow or deny on the phone

Cloud agentsChat with, schedule, and govern the agents that run remotely — Anthropic and AWS Bedrock alike.
+ New agent
All 11Acting for a person 8Workload 3
Name
Provider
Acts as
Status
Market Research AgentPrepares weekly market and rate briefings for the wealth advisory team.
Anthropic
The chatting userethan.park@highlandsbank.com
ActiveDisabled
Incident Summary AgentSummarised overnight fraud incidents for the morning stand-up. Kept for its records only.
Anthropic
The chatting userpriya.raman@highlandsbank.com
ActiveDisabled
Branch Operations AssistantAnswers branch staff questions about procedures and opens Service Desk tickets for facilities issues.
AWS Bedrock
The chatting userhannah.schultz@highlandsbank.com
ActiveDisabled

AgentPass · Cloud agents · Try it: flip a switch

Claude Code (Maya Thornbury)

claude-codeLiveWatched by Session Risk Reviewer

Whole session 09:25 – 10:58

Threads Root conversation and 2 subagents · width is share of requests

Root · 79%Explore · 10%general-purpose · 11%
Time
Activity
Thread
Tool
Target
09:25:00
File read
Root
Read
/Users/maya.thornbury/code/loan-portal/src/export/applicationExport.ts
09:25:57
Command
Root
Bash
git log --oneline -10
09:26:42
File read
Root
Read
/Users/maya.thornbury/code/loan-portal/src/export/applicationExport.ts
09:27:13
File read
Root
Read
/Users/maya.thornbury/code/loan-portal/src/api/rateLimit.test.ts
09:28:24
Search
Root
Grep
retryAfter
09:29:12
Network
Root
WebFetch
https://www.rfc-editor.org/rfc/rfc9110
09:29:56
Command
Root
Bash
git diff --stat
09:30:42
Network
Root
WebFetch
https://www.rfc-editor.org/rfc/rfc9110
09:31:17
File read
Root
Read
/Users/maya.thornbury/code/loan-portal/src/export/applicationExport.test.ts

AgentPass · Session · Try it: pick an activity

Your own AI is now the insider threat

Security can't say yes until it can answer four questions.

Agents act at machine speed on borrowed identities: a developer's API key, a shared service account, a static token. Most tools answer the first question. Almost none can tie an action back to a person.

01

What agents are running?

82% discovered shadow AI agents in the past yearCSA and Token Security

02

What are they allowed to do?

13% believe their AI agent governance is adequateGartner, 2026

03

Who is accountable for each one?

68% of organizations can't clearly tell agent activity from human activityCloud Security Alliance, 2026

04

Can we stop one right now?

21% have a formal process to decommission agentsCSA and Token Security

AgentPass in action

Pick a use case. Watch it run.

Every flow goes through the same gateway: identity, policy, approval and record, with the product screens your team will use.

Control unlocks autonomy

Break through the autonomy ceiling.

Agents create value by doing real work. Security allows that only when it stays in control.

Business value ↑
With AgentPassWithout AgentPass
With AgentPass24/7 agentic autonomySafe delegation across the enterprise.
Without AgentPassThe autonomy ceilingSecurity teams block or slow down agent work.
Agent autonomy

Start at your own pace

Start with Observability, then Expand from there.

AgentPass is SaaS and works with the identity provider you already run. No directory migration, and no change to how people sign in.

  1. 1Stage 1

    Get connected

    Console sign-in on your IdP, a first connector, and supervisor passkeys enrolled. First agents onboarded and a workstation cohort enrolled.

  2. 2Stage 2

    Observe

    Visibility into agent activity, with policies tested in Observe mode on live traffic.

  3. 3Stage 3

    Expand & Enforce

    Promote policies, turn on passkey approvals for high-risk actions, add teams and connectors.

Stages 1 and 2 · nothing is held or blockedStage 3 · enforce when ready

All you need to get startedAn identity provider, one connector, one or two supervisors with phones, and a few developer laptops.

Where existing approaches stop

Seeing agents isn't the same as stopping them.

AgentPass works alongside the security stack you already run. Here's the gap it fills.

AI gateways and LLM proxies

Where it stops

Route and log traffic. A laptop agent can go around them.

AgentPass adds

Keeps laptop agents on the governed path, with an owner on every request.

Identity providers

Where it stops

Issue agent identities, with no inline enforcement on each request.

AgentPass adds

Your IdP decides which apps an agent can connect to. AgentPass decides what it may do once connected.

Non-human identity platforms

Where it stops

Inventory accounts and rotate keys. No runtime policy on tool calls.

AgentPass adds

Policy on every call, a hold for a person, and a kill switch at request time.

AI security posture and DLP

Where it stops

Inspect content, not tool access.

AgentPass adds

Limits which tools and endpoints each agent can reach. Pairs well with content inspection.

Endpoint and browser security

Where it stops

Discover AI apps, then stop at blocking.

AgentPass adds

A governed path forward for legitimate use, so approved usage beats shadow AI.

Works with what you run
Coding agents
  • Claude Code
  • Codex
  • Cursor
  • and more
Cloud agents
  • Anthropic
  • AWS Bedrock AgentCore
Models and tools
  • Anthropic
  • OpenAI
  • AWS Bedrock
  • MCP servers
  • Your REST APIs
Identity
  • Microsoft Entra
  • Okta
  • Ping
  • Any OIDC provider
Built for the people who have to say yes

Four roles.
Four reasons to trust it.

FAQ

Questions security teams ask first

Our IdP already handles non-human identities. Why do we need AgentPass?

Your IdP stays the source of truth for people and roles, and AgentPass uses it. What an IdP doesn't do is enforce policy on each agent tool call, hold a request for approval, or stop an agent on its next request. AgentPass extends your IdP to agent runtime.

Won't human approval slow everything down?

Approval applies only where policy says so. Routine calls flow, and only named sensitive actions pause. Supervisors approve from their phone in seconds. We call this human above the loop: people set policy and step in on exceptions.

Does AgentPass stop prompt injection?

AgentPass doesn't scan prompts for injection. If an injection does succeed, the agent still can't call tools it hasn't been granted, and approvals are still held at the gateway, where the model can't reach them. Pair it with a content-inspection layer for detection.

Which platforms does the laptop component support?

Routing and egress blocking run on macOS 13 and later. Windows 11 reports agent activity today. It collects no prompts, model responses or file contents.

Do we need other HYPR products to use AgentPass?

No. AgentPass deploys independently and works with the identity and security stack you already run. It also integrates with the HYPR Identity Assurance Platform.

How long does it take to get started?

Start with one connector, one agent group and one supervisor group, in production or a sandbox. Developers install the HYPR app, enroll a passkey and point their agent at the gateway.

Get started

Your agents are already running.
So is the risk.

Start with one team: visibility first, then policies in Observe mode, then enforcement as you expand.

✓ Works with your IdP✓ No directory migration✓ Nothing enforced until you say so
Part of the HYPR Identity Assurance Platform

Built by the passkey company.

AgentPass brings HYPR identity assurance to AI agents, on the same FIDO2 passkey model HYPR runs across regulated enterprises. It deploys on its own and works with the stack you already run, whether or not it includes HYPR.