Agent-Native Authorization
Agent-Native Authorization
with Human-in-the-Loop
A framework that lets AI agents request just-in-time, human-approved authorization directly inside AI conversations and CLI tools, on open standards.
The Framework
Works for Every Type of Agent
TwoGenIdentity Agents
Our agentic orchestration layer, with JIT and Intent Authorization built in across the full agentic stack.
AI Assistants
GitHub Copilot, Claude and ChatGPT. The approval happens inside the conversation.
CLI Agents
Claude Code, GitHub Copilot CLI and custom scripts. The approval happens in the terminal.
How It Works
How Agent-Native Authorization Works
Five steps, the same for every agent type.
- 1
User Requests a Critical Action
The user asks the agent to perform a privileged operation.
- 2
Authorization Layer Enforces Protection
The gateway asks the AuthZEN policy decision point and requires step-up.
- 3
Agent Requests Human-in-the-Loop Input
The agent uses MCP Elicitation to ask the user inline.
- 4
User Completes Authentication
The user verifies with a passkey, OTP or authenticator app.
- 5
Elevated Token Issued, Action Succeeds
The identity provider issues a token bound to that operation, and the action completes.
Live Demo
See Agent-Native Authorization in Action
Try it yourself in IA+ Labs, or watch it running end to end.
Identity Access Plus
Managing a live Keycloak tenant with just-in-time authorization and passkeys
Also works with Claude Code and GitHub Copilot: OTP demo ↗ · Passkey demo ↗
Why Agent-Native Authorization
No Browser Redirect
Approvals run inside AI assistants and CLI tools.
Cryptographic Proof
Each elevated action gets a token bound to that exact operation, not a generic session.
SARC Authorization Model
Approvals use Subject, Action, Resource and Context, whatever the agent framework.
Bring human approval to your AI agents
Agent-Native Authorization is powered by the Identity Access Plus (IA+) platform.