The Japanese no-code site builder says an attacker reached its cloud environment on 30 September and 1 October and read the key that decrypts stored secrets. Japan's Digital Agency says its own site was affected.
Study finds 24,664 Lovable-built sites hand out database rows
EditorialBy TrustList Editorial
A Hamburg security firm found records with email addresses in 7,403 of them, from passport numbers to chat logs, and says the 30% share is not a confirmed leak rate.
About Study finds 24,664 Lovable-built sites hand out database rows
Study finds 24,664 Lovable-built sites hand out database rows
2 November 2026: Crissy Field, a Hamburg security firm working with the German business daily Handelsblatt, says 24,664 websites built with Lovable, the AI app builder, sent records from their Supabase databases to any visitor who loaded the page, with no login and no click. In 7,403 of them it found tables holding email addresses, and the records it pulled included passport numbers, contracts, invoices and private chat messages.
Not yet independently verified. The figures are Crissy Field's own and were not tested by us; its report carries no publication date, so we date it to Handelsblatt's first article of 2 November. We found no response from Lovable. We will update this when it can be confirmed, and remove this note.
How the sites were found
The researchers started from Common Crawl, the public web archive, and picked out 159,041 Lovable projects published between December 2025 and September 2026. Of those, 34,545 used Supabase as their database, and that group was the focus. For 24,664 of them, the browser received database records as soon as the page loaded. The team then made further read-only requests with the same public API key the site itself used, which is how it reached more content, and found email addresses in 10,560 tables across 7,403 sites, 30% of the group it examined closely.
What the records held
The report lists what it saw rather than naming sites. A German lifestyle forum exposed more than 80,000 profiles with full names, email addresses, dates of birth and gender, plus more than 200,000 chat messages. Ticketing and food-sales apps exposed buyers' emails, phone numbers, postal addresses, payment details and ticket codes. One property app exposed AI-written summaries of internal emails about tenants, owners and partners. Elsewhere there were more than 1,000 records with passport numbers, expiry dates and nationalities, tens of thousands of contracts, invoices, photo albums and internal WhatsApp messages, and 280,000 emails one provider had sent to its customers.
The cause, in the report's words, is access control mistakes that follow recurring patterns in how the apps connect to their databases. Because the patterns repeat, a misconfiguration found in one app is easy to look for in thousands of others.
The authors' own caveats
Crissy Field is careful about its headline number. Some of the email addresses may be public contact details or test data, while personal records without an email were not counted at all; the authors say the two effects partly cancel out but could not measure how far. "The 30% figure should therefore not be read as a confirmed data leak rate," the report says. The Common Crawl sample is not representative of every Lovable app, only parts of each dataset were read, and the team could not tell whether anyone had misused the data.
Handelsblatt put the findings to Lovable. The report notes that Lovable's terms of service place responsibility for an application and its data on the customer who built it.
Checking an app built this way
For a business that has let staff or an agency build internal tools or customer-facing apps on Lovable or a similar builder, the useful test is the one the researchers ran: open the app logged out, watch what the browser receives from the database, and try the same public key against other tables. The report's advice is to review access rules, interfaces and the separation between user accounts before an app handles real customer data, to repeat the review after significant changes, and to bring in outside help where the skills are missing. "Automated checks are no substitute for a professional security assessment," the authors write. Our no-code platform category lists 11 products.
Related on TrustList:
Sources
Categories & features
TrustList Weekly
The week in software and IT, in one email
The news that matters to buyers, new rankings and our own research. Every Thursday, free, and easy to leave.
You may be interested in
Anyone who knows a valid login ID can sign in without the password on Linux or Docker installs, and Windows with the Novell provider, against Active Directory or an LDAP server allowing anonymous bind.
Gems go read-only on 17 November and vanish on 17 December for personal accounts, with Workspace following in March 2027. Custom GPTs retire on 11 December, and custom actions do not survive the move to plugins.
Versions 22.23.5, 24.22.3 and 26.13.1 close HTTP/2 and TLS holes, including one that can leak TLS private keys from an HTTPS server's memory and one that bypasses client certificate checks.
CVE-2026-126187 scores 9.5 under CVSS 4.0 and needs no login. Four release lines get a fix, and apps that never render React on a server are not affected.
All seven are rated medium, but three can hand one user data cached for another, and one exposes HttpOnly session cookies on Edge routes to page scripts.
Versions 6.1.3, 6.0.10 and 5.2.19 carry the fixes. The one moderate issue is a GeoJSON request forgery, and full protection for another GIS flaw needs GDAL 3.9.
One critical Active Storage flaw allows arbitrary file reads and writes during image analysis, and four more are rated high. Security support for Rails 8.0 ends on 7 November, so 8.0 apps should plan the move to 8.1.
More on TrustList
Everything here links back to the same verified catalogue. Pick your next stop.
- More CybersecurityThe ranking for this subject
- CompaniesAgencies, consultancies and IT service providers, ranked by verified reviews.
- ProductsSoftware and SaaS with pricing, features, integrations and alternatives.
- AwardsAnnual recognition decided by verified reviews and an independent jury.
- LaunchesNew products and releases, voted up by the community every day.
- AI ModelsBenchmark scores and community ratings for every major model.
- RequestsBuyers describe what they need; vendors respond directly.
- PeopleReviewers, authors and makers with public profiles.
- ComparePut up to four listings side by side before you shortlist.