Built your app with AI? The security gaps we see most

AI-built apps · November 2026 · About a 7 minute read

AI app builders and coding assistants have changed who can build software. A founder with an idea can now launch a working app in a weekend. The catch is that these tools build what you ask for, and most people ask for features, not security.

The result is often an app that works perfectly for its users, and just as well for anyone who wants to read their data. Researchers who scan apps built this way keep finding the same problems. In one early study of apps made with a popular AI builder, around one in ten had flaws that let anyone read personal information. A larger scan of more than 5,000 publicly deployed AI-built apps found over 2,000 high-impact vulnerabilities and hundreds of exposed secret keys.

None of these problems show up when you use the app normally. That's why builders rarely know until a customer, a journalist or a criminal finds them first. Here are the five we see most, and how to check for each.

1. The database is open to everyone

Many AI-built apps store their data in a hosted database such as Supabase or Firebase. These are good products, but they rely on access rules (Supabase calls them row-level security) to decide who can read and change each record. If the rules are switched off, or written too loosely, anyone who finds your database address can read everything in it. And the address is usually sitting in your app's code, where every visitor's browser can see it.

How to check: open your database dashboard and confirm access rules are switched on for every table. Then ask: could a logged-out visitor, or a different logged-in user, read this table? If you're not sure, assume yes until someone tests it.

2. Secret keys in the code your visitors download

Your app probably talks to other services: payments, email, AI models, maps. Each needs a key. When an AI tool puts those keys in the front-end code, they're sent to every visitor's browser. Anyone can copy them and use your accounts, run up your bills or, with a payment key, see and refund transactions.

How to check: search your project for anything that looks like a key or password in front-end files. Secret keys belong on the server side, for example in a backend function or your platform's secrets store, never in the browser.

3. One user can see another user's records

This is the most common serious flaw in web apps generally, and AI builders make it easy to miss. A page loads an order, invoice or profile using a number in the address or request, such as /orders/1042. If the app doesn't check that the order belongs to the person asking, changing 1042 to 1043 shows someone else's data.

How to check: create two test accounts. Log in as one, note the address or request for one of your records, then log in as the other and try to open it. If it opens, you have a problem.

4. Nothing stops abuse

Rate limiting means capping how many requests one person can make in a short time. Without it, someone can try thousands of passwords, scrape your whole customer list, or hammer an AI feature until your bill explodes. AI-built apps very often have no rate limiting at all, because nobody asked for it.

How to check: find out whether your hosting platform or backend offers rate limiting, and switch it on for login, sign-up, password reset and any endpoint that costs you money.

5. The AI inside your app can be talked round

If your app includes an AI feature, such as a chatbot or an assistant that reads documents, it can be manipulated with cleverly worded input. That's called prompt injection, and it has topped the OWASP list of risks for AI applications in every edition. The danger grows with what the AI can access and do.

How to check: make sure the AI can only reach the data the current user is allowed to see, and that it can't take important actions, such as sending emails or changing records, without a person approving them.

A quick checklist before you launch

  • Database access rules are on for every table, and tested with two separate accounts
  • No secret keys in front-end code. They live in a backend or secrets store
  • Every page and request checks that the record belongs to the person asking
  • Rate limits on login, sign-up, password reset and paid features
  • Strong password rules and email verification on sign-up
  • Any AI feature can only reach the current user's data, and can't act without approval
  • Someone other than the builder has tested it

This isn't a reason to stop using AI builders

These tools are genuinely useful, and most of these problems are quick to fix once you know they're there. The fix is simply to treat security as a feature you ask for, and to have the result checked before real customers and real data arrive.

Want us to check your app? Our AI-Built App Security Review covers all of the above, with fix instructions you can paste straight into your AI builder. From $690.

Sources

  1. Wiz Research, Common security risks in vibe-coded apps
  2. The Next Web, coverage of security studies of AI-built apps, including the May 2025 study of 1,645 apps
  3. Summary of Escape.tech's scan of roughly 5,600 deployed AI-built apps
  4. OWASP API Security Top 10
  5. OWASP GenAI Security Project, Top 10 for LLM Applications