AI App Security Checklist & Readiness Score

The app works. Now ask the questions the AI may have skipped.

Use this before launch or before handing an AI-built app to real users. It focuses on the failures that are invisible in the UI: authorization, tenant isolation, secrets, webhooks, uploads, database policies and deployment permissions.

How this tool handles your dataRuns in your browser · no signup · your entered business data is not sent to NoustiqMethodology & limits

Authorization
Every data-changing and data-reading endpoint checks what the current user is allowed to access—not only whether they are logged in.
Weight: 10

Tenant isolation
You have tested that User/Tenant A cannot read or modify User/Tenant B data by changing IDs, URLs or request payloads.
Weight: 10

Database policies
Row-level security / equivalent database authorization is enabled and tested where the database is directly exposed through an API layer.
Weight: 9

Secrets
Production secrets, service-role keys and private API credentials are never shipped to browser JavaScript, public repositories or client-readable config.
Weight: 10

Payments & webhooks
Payment and other trusted webhooks verify signatures and reject replayed/forged requests.
Weight: 8

Input handling
Database queries are parameterized and untrusted input is validated/encoded for its destination context.
Weight: 8

File uploads
Uploads restrict size/type, are stored outside executable paths, use generated names, and are not trusted based only on extension or client MIME type.
Weight: 8

Abuse controls
Login, password reset, signup, expensive AI calls and sensitive APIs have server-side rate/abuse controls.
Weight: 6

Sessions
Session cookies/tokens use appropriate Secure, HttpOnly, SameSite, expiration and rotation/revocation controls.
Weight: 6

CORS & origins
Production CORS is restricted to required origins and is not broadly opened just to silence browser errors.
Weight: 5

Dependencies
Dependencies and lockfiles are reviewed/scanned, unused packages are removed, and critical security updates have an owner/process.
Weight: 5

Logging & recovery
Security-relevant actions are logged without storing secrets; backups, restore testing and incident shutdown/revocation procedures exist.
Weight: 5

Admin access
Admin/owner actions require server-side authorization and are not protected only by hidden routes or client-side UI checks.
Weight: 5

AI/agent boundaries
AI agents/tools do not receive broader filesystem, shell, cloud, database or production permissions than required.
Weight: 5

The app works. Now ask the questions the AI may have skipped.

Not assessed

This is a readiness checklist, not proof.

Get a deeper review before the blast radius grows.

Want an authorized review and remediation plan?