“Build an app in one prompt, no code needed” is the main character energy of 2026. But a new report says a lot of those one-prompt apps shipped with the database equivalent of leaving your front door open and posting the address on your story.
The receipts
German security firm Crissy Field, working with the business daily Handelsblatt, looked at 159,041 projects built with Lovable, the Swedish AI platform that turns plain-language descriptions into web apps. Per the report, 34,545 of them used the database service Supabase, and 24,664 sent database records to the browser as soon as the page loaded, no login needed.
The researchers are clear that this alone isn’t a leak: some data is meant to be public. But with extra read-only requests using the same key each site already hands to every visitor, they say they found records with email addresses in 10,560 tables across 7,403 sites (30% of the ones examined closely).
The examples they list are not a vibe:
- a lifestyle forum with 80,000+ profiles (names, emails, birth dates) and 200,000+ chat messages readable;
- event organizers and restaurants exposing tens of thousands of orders with phone numbers, addresses, payment information and ticket codes;
- a service provider exposing 1,000+ records with passport numbers, expiry dates and nationalities;
- contracts, invoices, photo albums, and in one case 280,000 emails sent to a provider’s customers.
OK but how does this even happen?
No hacking montage required. Per Cybernews, nobody forced anything: the researchers only used the access the sites already give every visitor. Here’s the mechanism in plain words:
- A Supabase app talks to its database straight from your browser, using a public key that is supposed to be public. Everyone gets it.
- What actually protects the data is Row Level Security (RLS): rules inside the database that say who can read which rows. Supabase’s own docs put it bluntly: a table in an exposed schema without RLS is readable and writable by any role with access to it.
- So if a table is missing those rules, or has a rule that basically says “everyone can read everything”, the public key becomes a backstage pass.
The report adds a spicy twist: because AI-built apps follow recurring patterns, a trick that works on one app may work on many others “with little extra effort”. Copy-paste architecture, copy-paste vulnerabilities.
The fine print (read before you dunk)
- Crissy Field says the 30% figure is not a confirmed leak rate: emails can be public contact addresses or test data, and the Common Crawl sample isn’t representative of all Lovable apps.
- The firm says it could not establish whether any data was misused, and it isn’t publishing how to access it.
- It didn’t examine Lovable’s own built-in database, and notes access-control mistakes happen in human-written apps too.
- Per Cybernews, Lovable responded that this is a known, industry-wide configuration problem around Supabase. The report also points out that Lovable’s terms put responsibility for apps and data on customers.
So the honest take: this is less “Lovable got hacked” and more “the no-code promise skipped the security tutorial”. The gap between “anyone can build it” and “anyone can read it” is one database setting.
How not to be the next case study
- Before real user data goes in, ask your AI builder to audit access rules for every table, then have a human who knows the stack check it. The report explicitly says automated checks don’t replace a professional review.
- Test as a stranger: log out and see what your app still shows.
- Don’t collect what you don’t need. Passport numbers in a side project? That’s a choice.
Prompting an AI to “make it secure” is not a spell. Prompting it to list every table, who can read it and why, then verifying the answer? That’s the skill. Practice it in the free Prompt Lab below. 👇
