DEV Community

Cover image for We scanned 50 apps built with Lovable and Bolt. 18 had a secret key in the code.
GuardCue
GuardCue

Posted on Edited on Originally published at guardcue.com

We scanned 50 apps built with Lovable and Bolt. 18 had a secret key in the code.

I build GuardCue, a checker for apps made with AI tools. Before telling anyone their app had a problem, I wanted to know how common these problems really are. So we pointed our own checker at 50 real, public apps that strangers built with Lovable and Bolt, and read the code of the 47 that were still public when we ran it on 30 October 2026.

No app is named here, and none ever will be.

What we found

What we found Apps
A key or password left in the code 18 of 47
A paid service (texts, AI calls) anyone could run up with no limit 17 of 47
Something a visitor types could reach a part of the app that treats it as an instruction 13 of 47
Outside code with a known, already fixed fault 32 of 47
Stored customer information readable by a stranger 1 of 6 we could check

A key in the code is like taping your shop key to the front door. Anyone who can read the code can copy it and use the app's accounts as if they were the owner.

A paid service with no limit means nothing stopped one person sending a hundred thousand AI requests overnight. The bill is how the owner would find out.

Why only 6 databases? To check this we need a working site to visit. Most of the 50 were code with no site running, or the site gave us no way to find where the data was kept. The other 44 are not "safe", we simply could not look.

How the 50 were chosen

We searched GitHub for public repositories, then confirmed each one by a file the tool itself writes: Lovable adds a package called lovable-tagger, and bolt.new writes .bolt/config.json. That matters, because the first ten Lovable search results included two awesome lists, which are not apps at all.

Forks and archived repositories were skipped, results were taken newest first, and no repository appears twice.

What we did to each one

We read a shallow copy of the public code once, then deleted it. Where an app had a live site, we took the public key the site already hands every visitor and asked each likely table for a row count. The answer is a number in a header, so it proves a table is readable without a single row of anybody's data reaching us. We changed nothing and signed in to nothing.

What these numbers do not say

  1. This is about the apps, not the tools. A key left in the code is something the person building it did, not something Lovable or Bolt wrote for them.
  2. When we checked a database, we were guessing table names, which is weaker than reading a real list.
  3. 50 apps is a sample, not a census.

If you built something with an AI tool

The three quickest things to look at yourself:

  1. Search your code for anything that starts with sk_, sk- or service_role. If it is there, treat it as leaked: make a new one and delete the old one.
  2. Put a daily limit on any paid service your app calls.
  3. If you use Supabase, open Table Editor and check every table shows RLS enabled with at least one policy.

The full method and the dataset (free to reuse under CC BY 4.0 with a credit) are on the original page: We scanned 50 apps built with AI tools.

What would you want a study like this to check next?

Top comments (0)