In the past, building a registration website might require hiring developers and waiting several weeks.

Now, with AI website building tools, you can launch a site in just a few hours by typing a few sentences.

But fast doesn’t mean secure.

In early November, German cybersecurity company Crissy Field released a study: they examined about 159,000 websites built with the AI website builder Lovable and found that 24,664 of these sites allowed anyone to read database content without logging in.

Some sites leaked over 1,000 passport records, others tens of thousands of event orders and tickets containing customer phone numbers, addresses, and payment information.

The researchers pointed out that the issue isn’t that the Lovable platform was hacked, but that database permissions behind the websites were not configured correctly.

This article uses a hypothetical case of a 3-person event studio to break down the key actions to take before launching an AI-built registration website.

1. What problem does this event studio face?

Imagine a 3-person event studio organizing 2-3 crafting classes and weekend markets per month.

Previously, registrations were done via forms plus LINE messages, with attendee lists scattered across multiple places—causing missed or duplicate registrations.

Owner Xiao Lin heard about AI website builders and spent one afternoon creating a registration site with event info, registration form, payment details, and a backend attendee list—all in one place.

The site looked great and registrations flowed smoothly.

But Xiao Lin hadn’t considered one question: where is the registrant data stored, and who can access it?

2. Why can AI-built websites sometimes allow “anyone” to access data?

Many AI website builders store data in cloud databases. For Lovable, the study showed it typically uses a database service called Supabase.

These databases have an important setting called “Row Level Security” (RLS), which basically controls who can read or modify which data.

If RLS isn’t enabled or isn’t properly configured, anyone with the “key” that the webpage naturally exposes could read the entire database.

Crissy Field’s researchers simply used the keys already provided to browsers, limiting themselves to read-only actions—and accessed a large amount of data.

The researchers also warned: “User-friendly development platforms do not replace expert checks of access permissions.”

It’s important to note that not all accessible data means a leak—some data might have been public info or test data. It’s also unclear if these exposed data sets have been misused.

3. Step one: Decide what NOT to collect

Xiao Lin’s original registration form asked for: name, phone, email, ID number, birthday, address, and last 5 digits of bank account.

After review, he found that for crafting classes and markets, only the following are actually needed:

  • Name: for check-in
  • Phone or Email: for notifications and contact
  • Event session and number of attendees registered

ID numbers, birthdays, and addresses aren’t necessary—never collect what you don’t need. The less data collected, the lower the risk if something goes wrong.

Also, don’t let customers enter bank account details on the form—instead, use formal payment methods with secure transaction flows.

4. Step two: Ask AI to check your database permissions

After building the site, Xiao Lin didn’t rush to launch. He asked the AI within the website tool:

Please check all database tables used by this website and verify if Row Level Security is enabled. List which tables and columns are readable by visitors without logging in. Can registrants see other registrants’ data? If any data is accessible without login, please explain the risks and suggest fixes.

After the AI’s report, Xiao Lin found out that the “registration list” table was readable by unauthenticated visitors. He then asked the AI to fix it so that:

  • Visitors can only “add” their own registration but cannot read anyone else's data.
  • Only logged-in studio admins can view the full attendee list.

However, AI claiming it’s “fixed” doesn’t guarantee it actually is, so there’s another step.

5. Step three: Test from a “stranger’s” perspective

Xiao Lin used another device, without logging in, to try:

  • Directly visiting the backend attendee list URL to see if access was blocked.
  • Submitting a registration with a test account and checking if other registrant data is visible.

The key here is: don’t just trust settings—test the site as an outsider would.

If you’re unsure of these steps or if your site handles payments or sensitive documents, it’s best to have a cybersecurity expert review before going live. That’s also the researchers’ recommendation.

6. Step four: Regularly delete old data after events

Registration data doesn’t need to be stored forever.

Xiao Lin set a rule: one month after each event ends, export the attendee list to a file stored in a secure location only he can access, then delete the old data from the website database.

This way, even if the site has a future issue, only recent limited data could potentially be exposed.

How long to keep data and how to handle it must comply with personal data protection laws—consult professionals if uncertain.

7. Roles: What AI does vs. what humans do in this process

  • AI is responsible for: building the website and registration form, checking database permissions, and suggesting fixes.
  • Humans are responsible for: deciding what data to collect, verifying AI’s fixes actually work, testing from an outsider’s view, and setting data retention policies.

AI makes “building websites” easier, but “protecting customer data” remains the business’s responsibility.

8. What are the limitations of this approach?

  • This is a hypothetical case; actual steps vary depending on your tool and database.
  • AI checks can miss issues and should not replace full cybersecurity audits.
  • The Crissy Field study focused on Lovable sites using Supabase and is not representative of all Lovable sites—but permission misconfigurations may exist in other AI builders too.

9. How to start testing on your own site

  1. Review your current registration form and remove all unnecessary fields.
  2. Ask AI to list what data is readable by non-logged-in visitors.
  3. Test on an unlogged-in device to see if you can access others’ info.
  4. Set a calendar reminder to clean old data one month after each event.

The most important takeaway today

AI can build your website in hours, but before going live, you must personally verify that only authorized people can see customer data.

Sources:

Recommended reading