Security Check for Apps Built with AI

Built it with Lovable, Bolt, Cursor, Replit or v0? Paste the address. In under a minute you'll see what a stranger can reach: leaked keys, database tables anyone can read, files that should never be public and API routes that hand out data without a sign-in. Every problem comes with a plain-English fix and a prompt to paste into your AI builder.

Read-only: it never writes, and never shows anyone's data, only table names and counts

Works on any public address. Checks the page, its JavaScript, your Supabase or Firebase rules, known file paths, the API routes your front end calls and the pages it links to.

Weekly Monitoring

Apps built with AI change every time you prompt them, and one change can switch off a security rule. Monitoring re-runs the check every week, keeps the history, and posts to your Slack or Discord the moment something new turns up, with the fix and the prompt in the message's link.

One-off check

Free
  • Everything above
  • Fix prompts for your AI builder
  • Smoke tests to download

Studio

$19 / month
  • 5 apps, checked weekly
  • For agencies and hackathon teams
  • Everything in Solo

What App Check Looks For

Data anyone can read

Supabase's anon key and Firebase's config are public by design: they sit in your front-end code. What protects your data is Row Level Security (Supabase) and security rules (Firebase). App builders often create tables without them, or with a rule that lets everyone in. In 2025 a researcher found more than 170 apps built with Lovable whose databases anyone could read (CVE-2025-48757). App Check uses your app's own public key, exactly as your app does, and reports:

Secret keys in front-end code

Everything in your site's JavaScript can be read by any visitor. App Check reads the page and the scripts it loads (and the chunks they load) for AI provider keys (OpenAI, Anthropic, Groq, OpenRouter, xAI), Stripe secret keys, AWS, GitHub and Slack tokens, database URLs with passwords, private keys and Supabase service-role keys. Keys are shown masked.

Missing sign-in checks

It finds the /api/… routes your front end calls and requests each one the way a stranger would, with a plain GET and no session. A route that answers with emails, names or tokens is flagged. Routes whose names suggest they change something (delete, pay, send, reset…) are never called.

Files that should never be public

.env files, the .git folder, .DS_Store and source maps that publish your original code. For .env only the variable names are shown, never the values.

Smoke checks

The home page and up to ten linked pages and app routes: status, speed and error output. From the results it writes a Playwright smoke-test suite and a GitHub Actions workflow you can drop into your repo, so every future change is checked in a real browser before your users see it.

The GitHub repo check

Reads Supabase migrations for tables without Row Level Security and policies that use using (true), Firebase rules in test mode, committed .env files, secrets with a public prefix (VITE_, NEXT_PUBLIC_), AI APIs called from the browser, API routes with no sign-in check, SQL built from strings, CORS that trusts every site, and whether the project has any tests.

What it can't see

Pages behind a sign-in, data rules that depend on who is signed in, business logic, and anything the front end doesn't reference. A clean result means none of the common holes were found from the outside, not that the app is secure. For apps that handle money or health data, get a human review too.

For coding agents and CI

The same scan is an API: POST https://www.phishclean.com/api/v1/scan-app with {"url": "…"}. See the developer docs. Building with Claude Code or Cursor? Check what your agent itself can reach with the MCP security audit.

Last updated: