Five things to check in any product built with AI
The short list we give every team whose product was built fast with AI tools, whether it's been live for years or launches next week. None of it needs a security team.
Your product works, customers rely on it, and a growing share of the code was written by Claude, Cursor or Codex. Here’s the list we give every team in that position, whether or not they hire us. None of it needs a security team. Most of it is an afternoon.
1. Put hard limits and alerts on anything that costs money
SMS, email, payments, AI model calls: every paid API you use. Set a daily cap and an alert well below it.
A stolen key or a bot shouldn’t be able to send 50,000 texts on your account overnight. If one does, you want a text at 2 a.m., not a bill at the end of the month.
2. Shrink your keys and lock the dashboards
Your app probably uses a full-access payment key because that’s what the tutorial used. Most providers let you create restricted keys with only the permissions your app needs. Swap them in.
Then turn on 2FA, ideally with a security key or passkey, on every provider dashboard: hosting, DNS, payments, auth, email, SMS and your code host. Those dashboards are where an attacker goes once they have a password.
3. Test whether one customer can see another’s data
This is the most common serious bug in AI-built apps, and it takes ten minutes to check:
- Make two test accounts, A and B.
- Log in as A. Click around and note the IDs in the URLs and API calls (orders, invoices, profiles, files).
- Log in as B and request A’s IDs directly.
If B can see or change anything that belongs to A, fix it before anything else.
4. Turn on logging for every route
Log the user ID and IP address on every request, not just the part of the app you were debugging last week. Keep the logs longer than your host’s default, which is often only a few days, and add an alert for a spike in failed logins.
When something odd happens, logs are the difference between “we know exactly what happened” and “we have no idea.”
5. Protect your main branch and put a check in front of deploys
Require changes to go through a pull request, and make your host deploy only after the checks pass. Then add a security review step to those checks. Your AI tools will keep writing code; a check that runs on every change catches the mistakes they repeat.
If you want a senior second set of eyes on the whole product, that’s what our Security Assessment is for.