01 · Overview
What security gaps do vibe-coded apps usually have?
The usual gaps are in who may do what, not in exotic attacks. The table lists the four areas, what to look for, and how a fix looks. They apply to apps from Lovable, Bolt, Replit, Cursor and similar tools, and they come up in any software project rescue.
| Area | What goes wrong | What a fix looks like |
|---|---|---|
| Authentication | Login works on screen, but the server does not check who is calling | Check the signed-in user on the server for every request that reads or changes data |
| Row-level security | A table is open, so any user can read every row | Enable RLS on every table, write policies per role, test with two accounts |
| Secrets in client code | An API key or service key ships to the browser | Move keys to server-side storage and rotate any that were exposed |
| Rate limits | Sign-up, login or paid API calls can be hammered | Set limits on auth endpoints and any route that costs you money |
02 · Authentication
How do you check authentication in an AI-built app?
Check that every route that reads or changes data verifies the caller on the server, not only in the interface.
- 01
AI tools often build a login screen that hides pages, while the underlying API answers anyone who knows the address.
- 02
Test it directly: sign out, call the data endpoints and confirm they refuse. Then sign in as user A and request user B's records. If you get them, the check is missing. Also look at password reset, email confirmation and any admin route, since these are built last and tested least.
03 · Data access
What is row-level security and why does it matter?
Row-level security is a database feature that decides, row by row, which users can read or change data.
- 01
Supabase's RLS guide says to enable it on every table in an exposed schema, and that once it is on, no data is reachable through the API with a publishable key until you create policies.
- 02
The guide also warns about three traps. A policy such as
to anon using (true)exposes every row to signed-out users. Views bypass RLS by default, so on Postgres 15 and later you setsecurity_invoker = trueto make them respect table policies. And on existing projects, new tables start with broad grants, so you should revoke them and grant back only what each role needs. Lovable's Supabase docs say missing policies are the most common way app data gets exposed.
04 · Secrets
Where do secrets leak in vibe-coded apps?
Secrets leak when a key is written into code that runs in the browser, or committed to the repository.
- 01
Anything shipped to the browser is readable by every visitor. Supabase's RLS guide says never to expose a secret key in the browser, because that key bypasses RLS when the request carries no user token.
- 02
Search the client bundle and the repository history for keys, tokens and connection strings. Replit's Secrets docs advise against hard-coding secrets and say they belong in the Secrets tool, where code reads them as environment variables. If a key was ever exposed, rotate it. Removing it from the code does not undo the exposure.
05 · Abuse
Why do rate limits matter for an AI-built app?
Rate limits matter because without them one script can fill your database with fake sign-ups or run up your bill on a paid API.
- 01
Set limits on sign-up, login, password reset and any endpoint that calls a paid service.
- 02
Some platforms provide part of this. Supabase's auth rate limits page says most auth limits can be changed in project settings, and that some are tracked per IP address and others per user. Check what your platform covers and add limits where it does not.
06 · Scope
Is a security review the same as a penetration test?
No.
- 01
A review reads the code, the data rules and the configuration and reports what is wrong. A penetration test attacks the running system from outside and is separate work, usually done by a specialist firm. We do the review and will tell you when a penetration test is worth adding.
- 02
Do not treat a clean review as proof the app is secure. It lowers the most common risks. Pair it with tests, so a fixed gap stays fixed. The prototype to production guide covers that.
07 · Next step
How does Ace8 review an AI-built app?
Ace8 is the modernization practice of AceMQ.
- 01
AceMQ engineers run a code audit on the app, check these four areas, and list what must change before launch. The audit is the first step of software project rescue. Price is a quote on request, with a fixed price agreed after a scoping call.
- 02
Request a rescue audit to get the findings before your users do.