1. Home
  2. Secure code review
  3. Is my vibe-coded app secure?

Is your vibe-coded app secure?

Probably not completely, and you cannot tell by clicking around. AI tools write code that works, not code that is checked for security. Here is how to find out where you stand.

The short answer is: probably not completely, and you cannot tell by using it. An app can look perfect and still let a stranger read every customer's data. Security flaws do not show up on screen. They show up when someone looks for them.

Why "it works" is not the same as "it is safe"

AI coding tools are built to produce code that runs and does what you asked. You asked for a login page, so you got a login page. You did not ask "what happens if someone skips the login page and calls the server directly?", so nothing was built for that.

The numbers back this up. One study from Carnegie Mellon found that roughly 10.5% of AI-generated code passes a security review. AI code is also reported to carry roughly 2.74 times more vulnerabilities than human-written code. These are not reasons to panic. They are reasons to check before you ship.

Five signs your app may not be secure

  • You pasted an API key into a prompt or a file. Hardcoded secrets and exposed API keys are the most common problem we see described in AI-built apps.
  • You never turned on row-level security. If your app uses Supabase and you did not write access policies, your data may be readable by anyone. See Supabase RLS and common AI-code mistakes.
  • You tested only as yourself. You never tried to open another user's page, an admin page, or an API route while logged out.
  • You never opened the package list. AI tools can suggest hallucinated dependencies, names of libraries that do not exist and that an attacker can register.
  • Nobody else has read the code. Not you, not a colleague, not a reviewer.

If one or more of these is true, the app has not been reviewed. That does not mean it is broken. It means nobody knows.

Three tests you can run in ten minutes

Test 1: the logged-out test

Log out. Open a page that should need a login and paste its address into the browser. If you see data, authorization is broken.

Test 2: the second-account test

Create two test accounts. While logged in as the first, copy the address of one of its records, log in as the second and paste the address. If the second user sees the first user's record, you have a missing authorization check.

Test 3: the source test

Open your live site, view the page source and search the downloaded scripts for words like secret, service_role or sk_. Anything found there is public.

These tests catch the obvious cases only. A code review goes further: a senior engineer reads your repository for secrets, your dependency list for packages that do not exist or are known to be vulnerable, and your configuration for missing access rules, then explains the result in plain language. It is a paid service, from 500 euro (indicative).

Ask about a code review

What happens if you ship anyway

The usual outcomes are not dramatic hacks. They are a leaked key that runs up a large bill, a customer who finds another customer's data, or a stranger who signs up as an admin. Each one costs trust, and some trigger a legal duty to report a data breach under GDPR if you handle personal data of people in the EU.

What to do next

  1. Run the three tests above.
  2. Work through the vibe coding security checklist, which covers secrets, authentication, injection and security headers.
  3. If you built with a specific tool, read securing apps built with Lovable, Cursor, Bolt or Replit for what to look at.
  4. If you want a person to read the code, ask about a code review to get a ranked list of what to fix.
  5. For anything that takes payments or stores personal data, book a human review. See how our code review works.

A scanner will give you a long list of warnings and leave you to sort out which matter. We read the report, explain it, and fix the problems in the paid review. You do not need to be technical to start. You only need to ask.

Common questions

Is my vibe-coded app secure?

Probably not fully. Apps built with AI tools usually work, but they often skip authorization checks, leave secrets in the code or ship database tables without access rules. The only way to know is to check.

Is AI-generated code safe?

It can be, but it needs review. One widely cited Carnegie Mellon study found roughly 10.5% of AI-generated code passes a security review, so assume your app has gaps until someone checks.

How can I tell if my app has been hacked?

Warning signs include unexpected bills from API or cloud providers, users you do not recognise, data that changed on its own and emails reporting leaked keys. Absence of signs does not prove safety, since many breaches go unnoticed.

Do I need a penetration test?

Not at the start. A code review plus a scan finds most problems in a young app at lower cost. A penetration test makes sense later, when you have customers or a compliance request.