Apps built with AI tools

AI-built app security audit

Your app was built with Lovable, Cursor, Bolt or Replit, and it works. We audit its AI-generated code and check whether a stranger can read your customers' data, change other users' records, make themselves admin or use your paid API keys at your cost.

From AED 15,000, excl. 5% VAT

Ask for a quote Book a call WhatsApp · See the sample report

How it happens

Most often through keys in the code sent to the browser, data anyone can read without logging in, and pages where one user sees another user's data.

What we check

  • Secrets and API keys in the code that the browser downloads.
  • Database and file storage rules: what a visitor can read or write without logging in.
  • Access between users and roles: whether one user can read or change another user's data.
  • Admin pages and API routes that check the role in the page only, not on the server.
  • Payment, webhook and file upload endpoints that trust what the browser sends.
  • Libraries with known flaws.

Ask for a quote

Tell us what you want tested. We will reply by email.

We use these details to answer you. They are stored as the privacy page explains.

What an automated AI review found in our lab

We ran Claude Code's /security-review three times on DVNA, a public web app written to be vulnerable, with 19 known flaws. It reported 10 of them in every run and 12 in its best run, with no false alarm. It never reported 7 of the 19. Among them are forms that another website can submit in a logged-in user's name, and a session secret written in the code.

Anthropic's own support page says such reviews "should complement, not replace, your existing security practices and manual code reviews". The full results are in the sample report.

What you receive

A written report. Each finding says what an attacker can do, how serious it is, the evidence, and how to fix it. The report also says what was tested and what was not.

See the sample report, built from our own lab tests.

Lab results, 11 Oct 2026 High

B-01 The automated review alone left 7 of the 19 known flaws unreported

What an attacker can do
Use any of the flaws B-14 to B-20, which the team was never told about.
Evidence
Reported in every run: 10 of 19.
Fix
Review by hand the classes that the tool's prompt excludes: outdated libraries, missing hardening, missing logs.
Retest result
Not applicable.

How it works

  1. A call to agree what is tested, and the price.
  2. You sign a written authorization. No test starts without it.
  3. We run the test.
  4. You receive the report.

Published work

Related