#129Case study3 min
Is your AI-built app really secure?
Look beyond a working demo to the code, data access and deployed behavior that need checking.
Hackaiz team · updated 21/11/2026

01A successful demo answers a different question
An application built with Lovable, Bolt or Replit can display the right pages and still give a user access to information that belongs to someone else. Functional testing asks whether the intended journey works. Security review also asks what happens when someone changes an identifier, calls an operation without the expected role, or uses credentials exposed in delivered code. The fact that an AI assistant generated the application does not settle those questions.
Consider an illustrative example, not a customer incident: a dashboard hides its administration menu from ordinary users, but the server does not enforce the same restriction. The screen looks correct while the underlying operation may remain accessible. Start by listing the sensitive actions in your app, such as reading private records or changing account permissions, and identify where those actions are enforced rather than only where their buttons appear.
02Match each question to the right evidence
A deployed website, a GitHub repository and a database reveal different parts of the application. A website review does not reveal every private source file or database access rule. Connect the authorized repository when you need source-level inspection, and the supported database connection when you need the relevant database controls examined. Connecting a system does not itself start an analysis, so review the intended checks before launching work.
For a Supabase-backed application, a useful review question is whether row-level security, the rules that limit which records a user can access, matches the intended tenant boundaries. Another is whether privileged credentials are kept out of browser-delivered code. These examples explain what to investigate, not a claim that your app contains those flaws or that one scan can prove every rule correct. Check the report’s actual scope before drawing a conclusion about stored data.
03Prove ownership before examining the live service
Before any website or API analysis, complete domain ownership verification. The interface offers “Via file (easiest with an AI builder)” and, where offered, “Via DNS (TXT)”. Publish the HTTPS file at /.well-known/hackaiz-verification.txt or the requested TXT record under _hackaiz. If you use a builder-managed address, ask the project’s AI to create the supplied proof file and publish the project; you may not control the provider’s DNS.
Use “Copy the instruction for my AI” when following that file-based path, then return to verify the proof. Do not paste private credentials into your builder chat to compensate for a failed verification. Once ownership is confirmed, select the checks you actually want. Unticked checks are not run or billed. Analysis uses credits; a missing permission or a check that did not finish must remain a visible limit, not be described as a passing test.
04Use the result to make a release decision
Read “Executive summary”, then inspect the findings under Danger, To fix and To watch. For each item, identify the evidence, the affected component and the suggested action. A report with no findings means no finding was produced by the checks that ran; it does not certify the whole app. If a material area was not examined, keep that gap in your release discussion rather than treating a reassuring headline as permission to ignore it.
When a code correction is eligible, Hackaiz proposes a draft GitHub pull request, a change request that still needs human approval and review. It never merges automatically. Test the change with the intended user roles, deploy it through your normal process, and verify the affected behavior again. Optional monitoring applies only to eligible connected systems and can support later follow-up; it does not block attacks. AI-assisted development can remain fast while release decisions stay grounded in evidence instead of confidence in the generator.
Frequently asked questions
Take the next step
Was this article helpful?