Control what AI systems
are allowed to execute

AI systems cannot execute sensitive actions without authorization.

AI System

procurement-agent

Action

Award Vendor Contract

$45,000

Authorizing

X|PLAIN

Decision

ESCALATE

Manager approval required

Allow
Block
Escalate

The Problem

AI is already executing actions

Refunds, payments, deployments, data access — without authorization controls.

Uncontrolled

AI Agent

Refund

$847 customer refund

Award Vendor

$45,000 contract

Deploy

Production environment

Delete

1,247 customer records

Payment

$12,500 vendor payment

No authorization layer

Most organizations cannot prove:

who authorized themunder which policywithin what scope

One unauthorized action. One audit finding. One production incident.

What It Looks Like

Authorization decisions

Every AI action is allowed, blocked, or escalated before execution.

Authorization DecisionsRecent activity
ActionAI SystemDecisionPolicyReasonTime
Issue Refundsupport-agentALLOWRefund PolicyWithin approved limits2s
Read Customer Datadata-agentBLOCKData Access PolicyOutside approved scope5s
Award Vendor Contractprocurement-agentESCALATEVendor Award PolicyManager approval required8s
Deploy Production Releaseci-agentBLOCKDeployment PolicyNot authorized12s
Send Customer Notificationnotify-agentALLOWNotification PolicyWithin approved limits15s
Access Analytics Dataanalytics-agentALLOWData Access PolicyWithin approved limits18s
Delete Customer Recordscleanup-agentBLOCKData Retention PolicyAction prohibited21s
Send Vendor Paymentbilling-agentESCALATEPayment PolicyApproval threshold exceeded24s
23 risky operations stopped12 actions awaiting approval842 actions executed safely
Last 15 minutes

After Deployment

What changes Monday morning

Four operational outcomes. Visible immediately.

Prevention

Your procurement agent cannot suddenly wire $100,000

Unauthorized actions never execute.

Fail-closed by default

Approval

High-value decisions require a human to say yes

High-risk actions wait for approval.

Built-in escalation workflows

Boundaries

Support bots stay in support. Procurement stays in procurement.

Each AI system stays within its approved role.

Per-system boundary enforcement

Accountability

When auditors ask who approved it, you'll have the answer

Every decision can be explained and defended.

Immutable decision records
<50ms p95 latency
99.9% uptime
Fail-closed by default

Integration

One line of code

Integrate once. Every AI action is authorized from that point forward.

authorize.ts

const decision = await xplain.authorize({

system: "procurement-agent",

action: "award_contract",

amount: 45000

})

// decision

ESCALATE

// Manager approval required

Your Application
Authorization Decision
<12ms
Decision
ESCALATE
ReasonManager approval required
Evidence
Recorded
XPLAIN

XPLAIN is an authorization layer for AI systems.

One authorization layer for every AI system instead of rebuilding approval logic inside each application.

Decision Reconstruction

When they ask who approved it, you'll have the answer

When auditors, regulators, investigators, or leadership ask what happened, you'll have the answer.

Action Request

Vendor Award

$45,000

AI System

procurement-agent

Policy

Vendor Award Policy v2.4

Reason

Manager approval required

Decision

ESCALATE

Waiting for manager approval

Evidence

Immutable record created

Timestamp

May 29, 2026 at 12:47:03 UTC

Audit-grade reconstruction

Before AI executes, XPLAIN decides.

AllowBlockEscalate
Request Demo
15-minute walkthroughReal authorization scenarios