AI & automationWebsites

How to Check Your AI-Built App Isn't Exposing Customer Data

Research out this week found records with email addresses open to anyone on 7,403 sites built with AI tool Lovable. How a small business checks its own AI app.

By Cybercave Technologies, LLC · Nov 3, 2026

Tablet with a keyboard case and a smartphone lying on bright pink, purple, yellow and teal paper — A tablet and a phone on colored paper. Illustration only, not an app or device from any company named here. Credit: Photo by Jakub Zerdzicki via Pexels (Pexels License).

Short answer: If you used an AI app builder to make a booking, order or signup app, check that its database shows each visitor only what they're meant to see. Research published this week found records with email addresses readable without a login on 7,403 sites built with Lovable. Start with the builder's security scan and a signed-out test.

Key points

Crissy Field, a security firm in Hamburg, Germany, reviewed 159,041 projects made with Lovable, an AI tool that turns plain-language requests into websites and apps. German business paper Handelsblatt reported the findings on Nov. 2.
On 7,403 of the 24,664 sites it looked at closely (30%), the firm could read database records containing email addresses without signing in. It says that figure is not a confirmed leak rate and that it could not tell whether anyone misused the data.
Lovable told Handelsblatt the problem is common across the industry. Under Lovable's terms, the customer is responsible for the app and how it handles data, according to Crissy Field and India Today.
The setting to check is the database's access rules. Supabase, the database service the study focused on, handles this with a feature called row level security.

What to do this week: open each app you built with AI in a private browser window, signed out, and make sure it shows no customer details. Then run the builder's security scan.

What did the researchers find?

They found customer and business records that anyone could pull up without an account. Crissy Field, which sells application security reviews, worked with Handelsblatt on the research and published its own report with updated figures.

The firm used Common Crawl, a public archive of the web, to find Lovable projects online between December 2025 and September 2026. Of the 159,041 it found, 34,545 used Supabase. On 24,664 of those, the page sent database records as soon as it loaded, which isn't a problem by itself if the data is meant to be public. With more read-only requests using the same key the page itself used, the firm found records with email addresses in 10,560 tables on 7,403 sites.

The report lists cases that hit close to home for small businesses. Several event organizers and restaurants had tens of thousands of records from selling tickets and food left open, including email addresses, phone numbers, mailing addresses, payment information and ticket codes. One service provider had more than 1,000 records with passport numbers. The report doesn't name the affected apps, and Crissy Field says it isn't publishing how it reached the data.

The firm is also clear about limits. Some of the addresses could be a business's public contact email or test entries, records without an email address weren't counted, and the sample may not represent every Lovable app. It looked only at apps using Supabase, not Lovable's own newer database feature.

What does Lovable say?

Handelsblatt reported that Lovable called this an industry-wide problem. The paper said it received tips about the issue in mid-August, checked them and put them to Lovable, and that its reporting shows the problem reaches beyond one company. India Today, citing Handelsblatt, reported the same response.

Lovable's own security documentation says its built-in scans can't guarantee complete security, that app owners are responsible for meeting the security needs of their use case, and that apps handling sensitive data may need a professional review. According to Handelsblatt, Lovable says more than 50 million projects have been built on it, with 1.2 million added each week.

Why can strangers see the data?

Because the app's database lets them. Apps like these often talk to their database straight from the visitor's browser, so the database itself has to decide who may read which records.

Think of a shop that locks its front door but leaves the stockroom door open to the alley. The page you designed may show only a thank-you message, but if the table behind it has no rules, anyone who asks the database directly may get every order.

Supabase's documentation says a table that apps can reach without row level security can be read and changed by any role allowed to use it, and tells developers to turn on row level security for every such table. It also warns that a rule letting signed-out visitors read every row should be used only for data meant to be public. Lovable's best-practices guide says it sets up basic rules automatically but that owners should review and adjust them early, before real data goes in.

Is my small business affected?

You should check if you, a staff member or a freelancer used an AI tool to build anything that collects customer information: a booking page, an order form, an event signup, a client portal or an internal tool with customer records. Handelsblatt reported the problem goes beyond a single provider, so the check applies to other AI-built apps too.

Our example (hypothetical): a Kalamazoo bakery used an AI app builder last year to take holiday preorders, with names, phone numbers and pickup times. The page looks fine to customers. Nobody checked whether the orders table can be read by someone who isn't signed in.

How do I check my AI-built app?

Most of this doesn't need coding. If anything looks wrong, stop collecting new data in that app until it's fixed.

List what you've built. Write down every AI-built app or form and what customer information it stores.
Run the builder's security scan. In Lovable, a quick scan runs each time you publish, and you can run a deeper scan from the project's Security view or ask for one in the project chat, according to Lovable's documentation. Fix anything marked Critical first.
Check the database rules. Every table with customer information should have row level security turned on, and no rule should let everyone read it unless the data is meant to be public.
Test it from the outside. Open the app in a private window without signing in. Then sign in with a second test account and make sure it can't see the first account's records.
Collect less. Our advice: don't ask for ID numbers or card details in a homemade app. Keep only what you need for the job.
Get a second set of eyes. Crissy Field and Lovable both say automated scans don't replace a review by someone who knows security, especially before an app handles sensitive data.

For your developer: confirm row level security is enabled on every table in an exposed schema (Supabase), review grants for the anon and authenticated roles, remove any policy that grants anon read access with using (true) on non-public tables, check views (they bypass row level security by default unless set otherwise), keep the service role key server-side only, and run Lovable's Deep scan. Repeat after adding tables or changing sign-in.

What should you do this week?

Pick the AI-built app that holds the most customer information and test it signed out in a private window. If you can see anyone's details, stop taking new entries, fix the database rules, and talk to whoever built it about what was exposed and for how long.

For more, see our guides to capping your bill if an AI key gets exposed, checking the third-party scripts on your website, and giving your AI assistant its own account.

FAQ

Was Lovable hacked? The report doesn't describe a break-in at Lovable. Crissy Field says it reached the data with read-only requests to the apps' databases, and it describes the problem as access-control mistakes and misconfigurations.

Do I have to be a Lovable customer to be at risk? No. The study only covered Lovable apps using Supabase, but Handelsblatt reported the problem isn't limited to one provider. Any app that lets visitors' browsers talk to a database needs the same rules.

Was my data in this study? Crissy Field hasn't named the affected apps, and it couldn't tell whether anyone misused the data. If you gave information to an app built this way, you can ask the business that runs it how the data is protected.

What if I find customer data exposed? Close the gap first, then write down what was exposed and for how long. Talk to an attorney about whether you have to notify anyone.

Disclosure: No company or organization named in this article paid for it, reviewed it or saw it before publication. Cybercave Technologies builds, hosts and manages websites for small businesses and offers AI automation services. Study figures are as published by Crissy Field and were checked on Nov. 3, 2026. Not sponsored.