Skip to content

AI-generated app security audit

An AI-generated app security audit finds the holes that AI coding tools leave behind, at a fixed price, each one proven with a request you can replay. Built for founders and teams who shipped a product mostly built with AI and now have real users and real data.

What is an AI-generated app security audit?

An AI-generated app security audit is a penetration test for a product that was built mostly by AI coding tools. Grapefruit attacks your live app and its APIs the way a real adversary would, finds the insecure patterns those tools tend to repeat, and hands you the exact request that proves each one.

You shipped fast. The app works, users signed up, data is flowing. The audit answers the next question: can one of those users read another's data, and can anyone on the internet reach something they should not. You get the answer with proof, not a maybe, and a fixed price so the audit itself holds no surprises.

Why do apps built with AI share the same holes?

Apps built with AI share the same holes because the tools learned from the same public code, including its mistakes, and they reproduce those mistakes at speed. Vibe coding ships a feature in minutes and ships the missing authorization check with it. The same handful of flaws show up again and again, which is exactly why they are fast to find.

The model writes the happy path. It wires the button to the endpoint, returns the data, moves on. The checks that a careful engineer adds out of habit, who owns this record, should this caller see it, how often can this be called, are the parts most likely to be skipped, because nothing in the prompt asked for them.

  • The endpoint returns a record by ID and never checks who owns it.
  • An API key or database secret sits in the code, committed to the repo or shipped to the browser.
  • A login or signup route has no rate limit, so it can be hammered all day.
  • Cross-origin rules are wide open, so any website can read a signed-in user's data.
  • User input is pasted straight into a query or a shell command without sanitizing.
  • Error messages print a stack trace or the raw database query, handing an attacker a map of your internals and sometimes the data itself.
  • A debug or admin route from early development is still live in production.

Which flaws show up most in AI-built apps?

The flaws that show up most are a short, familiar list, and each one has a clear payoff for an attacker. Grapefruit tests for all of them on every route and every endpoint your app exposes. The table pairs the common flaw with what an attacker walks away with when it is there.

Common flaw in AI-built appsWhat an attacker gets
Missing authorization check on a recordReads or edits another user's data by changing one ID
Secret or API key committed or shipped to the browserYour keys, your bill, and often a path straight into your database
Endpoint with no rate limitPassword guessing, data scraping, and runaway cost on your bill
Overly open CORSA malicious site reads a logged-in user's data from their browser
Unsanitized input into SQL or a shell commandThe whole database, or code running on your server
Debug or admin endpoint left liveInternal config, other users' records, sometimes admin actions

What do these flaws look like in a real app?

In a real app these flaws are simple to describe and expensive to suffer. The attacker already has an account, watches the traffic their own browser sends, and edits one value. Nothing exotic is involved, which is the point: the holes are ordinary, so finding them is a matter of looking in the right places quickly.

  • The borrowed record. A request loads document 1041. The attacker asks for 1042. The server checks that the caller is logged in, not that the document is theirs, and returns a stranger's file.
  • The key in the bundle. The app calls a paid service from the browser with the API key right there in the JavaScript. Anyone who opens dev tools copies the key and spends on your account.
  • The open door. A signup endpoint accepts unlimited attempts. A script tries thousands of email and password pairs, finds the accounts that reuse a leaked password, and logs in as them.
  • The field nobody guarded. A profile update saves every field it receives. The attacker adds a field that sets their role to admin, and a trial user now runs the workspace.

Why is a replayable request better than a scanner report?

A replayable request is better because it removes doubt. A scanner flags a maybe and leaves you to guess. Grapefruit gives your engineer the exact call that worked and the response that came back, so they paste it, run it, and watch another user's data land on their own screen. After the fix, the same call returns a refusal. That is the difference between a ticket your team argues about and one they close.

Buyers ask for this. In a Pentest-Tools survey of 158 respondents in June 2026, 53% ranked proof of exploit as a top buying factor, and 87.8% said they hit AI pentest findings that needed significant manual validation. A request you can run is the validation, done already, and it doubles as the regression test for the fix.

Do you test the web app and the API together?

Yes, in one test, because that is how the attacks actually chain. The front end hides a button, the API still answers the call behind it. Grapefruit drives your app through its pages and forms, then talks straight to your REST and GraphQL endpoints, including the ones no page ever calls, and reports it all in one place.

This matters doubly for AI-built apps, where the front end and the backend were often generated in separate passes. The page looks locked down while the endpoint behind it trusts anyone who is signed in. Authenticated testing with the test accounts you supply proves which one is true, end to end.

How fast can you start, and what does it cost?

You can start today. Add your hostname, publish one DNS TXT record to verify it, sign the testing authorization online, and testing begins. The first results land within hours of kickoff. You are not buying a slot on a calendar weeks out, you are buying results that start the same day. A full audit is a fixed quote, so you know the price before you commit, and maintenance plans re-test your verified domains every week or month as you keep shipping.

For reference, traditional web application pentests commonly cost $5,000 to $30,000 and take 2 to 4 weeks just to start, according to the Astra cost guide and Stingrai. See pricing for the plans, or open the sample report to see what a finding looks like before you buy.

Questions

My app was built with AI coding tools. Is it less secure?
Not automatically, but it tends to carry a predictable set of holes. AI coding tools reproduce the insecure patterns they were trained on, so missing authorization checks, exposed keys and open endpoints show up often. The upside is that the same patterns are fast to find. An audit confirms which ones are in your app and proves each with a request you can run.
What do you need from me to start?
Your hostname, one DNS TXT record to verify you own it, an online signature on the testing authorization, and test accounts. Accounts in two roles and, if you have tenants, two tenants give the best coverage, because the most damaging flaws in AI-built apps are one user reaching another user's data or a normal user reaching admin actions.
Will testing break my production app or leak real user data?
Point the test at staging or a dedicated test tenant and every action stays there. The test uses real exploitation through the accounts you supply, so it proves a flaw against data you control rather than your customers' records. You decide the target before anything runs.
How is this different from running a free scanner?
A scanner matches patterns and reports maybes you still have to check. Grapefruit exploits the flaw for real and hands you the exact request and response that prove it, ranked critical, high, medium or low in a portal. Your engineer replays the request to see the attack work, then replays it after the fix to confirm it fails.
I keep shipping new features with AI. How do I stay secure?
A maintenance plan re-tests your verified domains every week or every month. Each AI-generated feature you ship gets attacked on the next run, and each dated run shows what is new and what got fixed since the last one. That keeps pace with vibe coding, where a new endpoint can land any day.
How much does an audit cost and how long does it take?
A full audit is a fixed quote, priced before you start, and maintenance plans have public prices with card checkout. There is no demo call or scheduling step. Setup takes a few minutes and results land within hours of kickoff, rather than the 2 to 4 weeks a traditional engagement commonly takes to even begin. See pricing.

Standards and sources