New kiosk of pale wood and glass with an open, cyan-glowing door, a pulled-out metal drawer and a pile of blank cards and envelopes on a dark concrete floor (AI image)

Illustration

Developments3 min read

Lovable study: AI builds the app, and then nobody runs it

A Hamburg security firm scanned nearly 160,000 Lovable projects and found exposed records on thousands of sites, from tickets to passport numbers. The apps hit hardest are the ones marketing teams spin up themselves: sign-ups, waitlists, newsletters.

Key facts

  1. Crissy Field analyzed 159,041 Lovable projects from Common Crawl (December 2025 to September 2026); 34,545 of them use Supabase.
  2. 24,664 sites send database records on page load without a login; on 7,403 the tables contained email addresses (not a confirmed leak rate, per the researchers).
  3. Exposed data included more than 200,000 chat messages from one forum, ticket and order data with payment details, and more than 1,000 records with passport numbers.

Manfred „Luigi“ LugmayrBy Manfred „Luigi“ Lugmayr

Researched and written with AI, to Luigi's brief, positions and editorial rules. Automatically checked against the sources. How I work →

You know the drill. The event is three weeks out, IT has no bandwidth, so someone on the team builds the sign-up page themselves. In Lovable, in an afternoon. Tickets, waitlist, newsletter checkbox. Ship it. Everyone's happy. And then nobody ever asks who actually runs the thing.

Those are exactly the apps Hamburg security firm Crissy Field looked at, in a joint investigation with German business daily Handelsblatt. The English version of the report is sitting high on Hacker News today with more than 900 points.

What the study found

The researchers analyzed 159,041 Lovable projects found in the Common Crawl web archive (December 2025 to September 2026). 34,545 use the Supabase database. 24,664 of those sites send database records to the browser on page load, no login required. On 7,403 sites the tables contained email addresses. The researchers are explicit that this roughly 30 percent is not a confirmed leak rate.

The examples are still brutal. One forum with more than 80,000 profiles and more than 200,000 chat messages. Tens of thousands of ticket and food orders from events and restaurants, with addresses, phone numbers and payment details. More than 1,000 records with passport numbers at one service provider. Plus contracts, invoices, newsletter lists, waitlists and support requests. The researchers trace it to access-control mistakes in the database integrations, following recurring patterns. The report includes no response from Lovable.

This isn't a tool problem. It's an ownership problem

Look at that list again: tickets, waitlists, newsletters, support requests. Those aren't IT projects. They're marketing assets. Landing pages with a database behind them, built by exactly the people vibe coding was made for: fast, no IT ticket.

To me that's the real story. AI made building so cheap that it no longer triggers an approval. A campaign app used to be a project with a budget, a brief and a sign-off. Now it's an afternoon. The work before launch shrank. The responsibility after launch didn't. AI doesn't fix broken processes, it widens the cracks: wherever there was never a sign-off process for customer data, there are now hundreds of small apps with no owner.

And customer data is the one thing I'd never let run on the side. Briefs, copy, design: hand them to AI and freelancers. Who can access which data, and who gives the final approval, stays in-house, with a name attached.

The catch

This isn't a verdict on every Lovable app. The sample comes from a web archive, and exposed records don't automatically mean anyone misused them. The researchers themselves say automated checks are no substitute for a professional security assessment. If you calm yourself down with a scanner, you've just moved the problem.

Same pattern as the Cursor breach via Fireworks: the shortcut is convenient until someone finds the back door.

What I'd do now

  • Inventory this week: Ask your team which pages and apps were built with Lovable, Bolt, Replit and the like and store customer data. Event sign-ups and waitlists first.
  • One owner per app: a person, not a department. Whoever shipped it owns it until it's switched off.
  • Sign-off before real data: No self-built app collects customer data until someone with security know-how has reviewed the access rights. If you don't have that in-house, bring it in.

Bottom lineAI builds the app in an hour. Who runs it afterwards is your call, and you make it before the first customer email lands in the database.

  • Share

Got AI news? Submit news →

FAQ

Was Lovable itself hacked?

No, the study doesn't say that. The researchers trace the exposed records to access-control mistakes in the database integrations of individual apps.

Does this mean 30 percent of all Lovable apps leak data?

No. The sample comes from a web archive, and the researchers state explicitly that the roughly 30 percent is not a confirmed leak rate.

Source

Crissy Field: Lovable: personal data readable without login (research report)

More sources

  • Techzine – 4.11.2026, bestätigt Zahlen; Ursache schlecht konfigurierte Supabase-Datenbanken, keine Lovable-Stellungnahme im Artikel.
  • Hacker News – Nur Hinweisgeber: Bericht am 15.11.2026 auf der Startseite (938 Punkte).

Black Friday 2026

Black Friday deals on AI tools, straight to your inbox.

I collect the AI tool deals worth having and send them in the AI Briefing, along with the AI news that matters, every Wednesday and Friday. Five minutes to read.

  • Black Friday deals on AI tools, collected for you
  • The key AI news, briefly explained
  • Wednesday and Friday, unsubscribe in one click

In the AI Briefing right now

Image for the story: Perplexity and Runway: Black Friday brings credits, not discountsDealPerplexity and Runway: Black Friday brings credits, not discounts
Image for the story: AT&T makes its AI agents earn their autonomyNewsAT&T makes its AI agents earn their autonomy
Image for the story: North Korea's deepfake meetings: a face on the call proves nothing nowNewsNorth Korea's deepfake meetings: a face on the call proves nothing now
Blonde woman with glasses smiling as she scrolls on her phone in an office

Manfred “Luigi” LugmayrFrom Manfred “Luigi” Lugmayr: 30 years in marketing, seven of them as CMO at burgerme.