A laptop displaying code on a wooden desk, in a dimly lit workspace

Lovable data exposure: 24,664 apps served database records with no login, researchers say

By Yash Malviya·

Photo: Daniil Komov / Pexels

Crissy Field scanned 159,041 Lovable projects and found 24,664 sites that sent Supabase records on page load. The sample is not representative, but the exposed data includes chat messages and passport numbers.

Security researchers say thousands of web apps built with the AI app builder Lovable were handing out database records to anyone who asked. The firm Crissy Field scanned 159,041 Lovable projects and found 24,664 sites that sent records from a Supabase database as soon as the page loaded, with no login. The finding has circulated since early November 2026. The numbers need careful reading, because the authors themselves warn against treating them as a leak rate.

What the researchers did

Crissy Field's report says it used Common Crawl, a public web archive, to identify 159,041 Lovable projects built between December 2025 and September 2026. Of those, 34,545 used Supabase, a hosted Postgres database service. The team focused on Supabase because its integration is easier to analyze and is believed to be the most widely used database service among Lovable projects. It did not examine Lovable's own database features.

On 24,664 websites, the page sent Supabase database records on load without a login. Further read-only requests using the same key that the page gives every browser retrieved more content. That is the point: the key was never secret. The question is whether the database behind it checks who is asking.

The report says records containing email addresses appeared in 10,560 tables across 7,403 of those 24,664 sites, or 30 percent. It then lists its own limits. The email addresses may be public contact addresses or test data. Records without emails were missed. The 30 percent is not a confirmed leak rate, the sample is not representative, only parts of the datasets were retrieved and access details are withheld.

What was exposed

The report describes a German lifestyle forum with over 80,000 profiles and over 200,000 chat messages. It describes event organizers and restaurants whose tens of thousands of ticket and food sales records included emails, phone numbers, postal addresses, payment information and ticket codes. A property management company's AI-generated summaries of internal emails were readable. One service provider's database held over 1,000 records with passport numbers, expiration dates and nationalities.

The list goes on: tens of thousands of contracts and invoices, personal photo albums, internal WhatsApp communications, newsletter subscriptions, waiting lists, support inquiries and 280,000 emails sent to a provider's customers. These descriptions are the researchers' own. We have not independently inspected any of the data, and the report withholds access details.

“Automated checks are no substitute for a professional security assessment.”

Crissy Field, Lovable research report, November 2026
A woman with dreadlocks typing on a laptop in a dimly lit room at night
A developer at a laptop; AI app builders let non-specialists ship databases whose access rules they may never see. Photo: SHVETS production / Pexels

Who reported it, and what is missing

Handelsblatt published a report by Lina Knees on November 2, 2026, per Crissy Field's page, and Techzine, in a piece by Berry Zwets dated November 4, summarized the research. Techzine says the cause is poorly configured Supabase databases and that the researchers believe the same integration patterns recur across Lovable apps, so one attack could work on others with little extra effort.

What we could not find is Lovable's own response. The Crissy Field page contains no Lovable statement, no Supabase statement and no disclosure timeline. The Techzine article, as we read it, carries none either. We have therefore not quoted either company. Readers should look for a primary statement from Lovable and Supabase before drawing conclusions about who is responsible.

Why this is an AI story

AI app builders let people ship working software without writing or reviewing code. That is their appeal and their risk. A tool can generate a login screen and a database connection that look finished while leaving the database open. The person who prompted the app may not know that the access rules, not the interface, decide who sees the data.

Supabase's own documentation is direct about the mechanism. It says "A table in an exposed schema without RLS is readable and writable by any role with a grant on it." RLS stands for row level security, the setting that makes the database check who is asking for each row. Once it is enabled, the docs say, no data is accessible through the API using a publishable key until policies are created. We explain how this works for non-specialists in our explainer on row level security.

That is a configuration step, not a new vulnerability class. It is also exactly the kind of step that generated apps skip, and nobody notices because the app looks and behaves normally to its owner.

What operators should do now

Anyone running an AI-built app with a Supabase backend can check three things. Confirm row level security is enabled on every table in an exposed schema. Look for policies that grant access to every visitor, since the docs warn that a policy of to anon using (true) exposes every row to anyone unauthenticated. And never place a secret key in the browser, because, as the docs put it, secret keys bypass RLS.

If an app has held personal data in an open table, treat that as a potential data exposure with legal duties that vary by jurisdiction, and get professional advice. Crissy Field's own conclusion applies to its scan and to any quick fix: "Automated checks are no substitute for a professional security assessment."

Our take

This is a configuration failure at scale, surfaced by a scan that its authors describe honestly and cautiously. The 24,664 figure is a count of sites that answered a query, not proof of 24,664 breaches. The lesson for builders is to treat a generated app as unreviewed code. The lesson for the platforms is to make the safe configuration the default. We will watch for statements from Lovable and Supabase, and for any regulator response.

Frequently asked questions

Did Lovable leak user data?

Researchers at Crissy Field say 24,664 Lovable-built sites in their scan sent Supabase database records without a login. They describe a configuration problem in individual apps, and they say the figures are not a confirmed leak rate. We found no Lovable statement in the sources we opened.

How many Lovable apps were exposed?

Crissy Field identified 159,041 Lovable projects, of which 34,545 used Supabase. On 24,664 sites, database records were sent on page load without authentication. The authors say the sample is not representative, so the number does not describe the whole platform.

What data was exposed in the Lovable research?

The report describes over 200,000 chat messages from a forum, ticket and payment records, internal emails summaries, and over 1,000 records with passport numbers, expiration dates and nationalities. These are the researchers' descriptions, and access details are withheld.

What is row level security in Supabase?

Row level security is a Postgres feature that makes the database check, for each row, whether the requester may see it. Supabase docs say a table in an exposed schema without it is readable and writable by any role with a grant on it.

How can I check if my AI-built app is exposed?

Confirm row level security is enabled on every table in an exposed schema, review policies that allow all visitors, and keep secret keys out of the browser. Crissy Field warns automated checks are no substitute for a professional security assessment.

Sources

What each one is, and whose it is.

  1. 1

    Lovable research report, Crissy Field (November 3, 2026)

    OtherIndependent of the vendor
  2. 2

    Thousands of Lovable apps are leaking personal data, Techzine Global (November 4, 2026)

    Press reportIndependent of the vendor
  3. 3

    Row Level Security, Supabase documentation (November 16, 2026)

    Documentation