Skip to content

Launching and running it

What security checks should I make before sharing my AI-built app?

Lumos · Updated

Start with what the app stores and what someone could do without permission. Check private records, administrator actions, and any service credentials the app uses. Ask for evidence that the controls work before inviting people to trust the app with their information.

Checks to work through

  • Use separate test accounts to check who can read, change, and delete a record. Test signed-out access too. OWASP's authorization guidance calls for permission checks on every request.
  • Have the implementation reviewed for private keys in browser code, public files, or logs. Store credentials through the host's secret-management facility. Follow OWASP's secrets guidance when handling or replacing exposed credentials.
  • Check that invalid submissions are rejected by the server. Browser form rules help users, but they can be bypassed, as OWASP's input-validation guide explains.
  • Review account recovery, administrator access, and limits on actions that send messages or consume paid services. Check how you'll notice errors and restore saved data if something goes wrong.

Ask the reviewer or AI to identify what it checked, what failed, and what remains unverified. An automated scan or a successful sign-in doesn't establish that the whole app is secure.

Use disposable sample data while these checks are incomplete. If the product handles sensitive records or important business operations, get an experienced developer or security specialist to review the relevant controls. OWASP ASVS can help define that review's scope.

In the Builders Handbook

Make it work with real data introduces access rules. Test it, fix it, and keep going explains checking failures and recovery.