Thousands of Lovable apps expose passports, chats and payment data without a login
Thousands of apps built with the AI platform Lovable let strangers look into their databases without signing in. That's the finding of an investigation by German security firm Crissy Field together with Handelsblatt. Among the exposed data: passport numbers, payment details and hundreds of thousands of private messages.

What's going on
Lovable is one of the best-known platforms for so-called vibe coding. You describe the app you want in plain language, and an AI writes the code, hooks up a database and puts the app online. According to Lovable's marketing, you don't need to know how to code. That's now turning into a problem: thousands of apps built this way let strangers look into their databases without signing in.
That's the finding of an investigation by German security firm Crissy Field, carried out together with the business daily Handelsblatt. According to Startup Fortune, it was first reported on November 2. In the openly accessible databases, the researchers found passport numbers, payment information, valid ticket codes and private chat messages. Crissy Field couldn't determine whether anyone has misused the data. The company isn't naming the affected apps or explaining exactly how it got access, so nobody can exploit the holes deliberately.
The numbers
The researchers searched Common Crawl, a freely available archive that regularly saves large parts of the internet. There they found 159,041 Lovable projects that were online between December 2025 and September 2026. Here's how the numbers narrowed down:
| Step | Count |
|---|---|
| Lovable projects found | 159,041 |
| Of those, using the Supabase database service | 34,545 |
| Sending records on page load without a login | 24,664 |
| Websites with readable email records | 7,403 |
| Affected tables containing email addresses | 10,560 |
A site sending data on page load isn't a mistake in itself. A restaurant wants its menu to be public, after all. The trouble started with the next steps. The researchers sent more read-only requests using the same key the site hands to every visitor anyway, and that got them a lot more content. On 7,403 websites, or 30 percent of the sites they examined more closely, they found tables containing email addresses.
Crissy Field adds its own caveat: the 30 percent is not a confirmed leak rate. Some of the email addresses are public contact addresses or test data. The Common Crawl sample also doesn't represent all Lovable apps.
What was exposed
The examples in the report show how sensitive some of the data is:
- ID documents: One service provider exposed more than 1,000 records with passport numbers, expiry dates and nationalities.
- Community: 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.
- Tickets and food: Several event organizers and restaurants exposed tens of thousands of orders, complete with phone numbers, addresses, payment information and valid ticket codes.
- Property management: A real estate company exposed AI-generated summaries of internal emails about tenants, owners and business partners.
On top of that came tens of thousands of contracts, invoices, private photo albums and internal WhatsApp threads. In one case, 280,000 emails to a provider's customers were readable. According to Startup Fortune, this was a CRM-style project, meaning customer management software. Waitlist entries and support tickets were exposed there as well.
Where the hole is
Nobody had to break into Lovable or Supabase. Supabase is a database service that Lovable connects to apps by default. Apps that use Supabase send a key to the browser, the so-called "anon key". That key is public on purpose. The real protection comes from access rules inside the database, called Row Level Security (RLS). These rules decide, row by row, who gets to see which record, for example only your own orders. Think of an apartment building: the front door is open to everyone, and only the apartment doors are locked. If those locks are missing or set too loosely, anyone with the key can read every table.
RLS has to be switched on and configured correctly. Startup Fortune reads the findings this way: unless you specifically tell it to, the AI agent builds working apps without fitting those locks. At the same time, Supabase gives the public key broad read access as long as nobody restricts it.
According to Crissy Field, Lovable's database connections follow recurring patterns. So if you find a misconfiguration in one app, you can scan thousands of other apps for the same mistake with little effort. Startup Fortune puts it like this: one forgotten checkbox is a bug. When the same forgotten checkbox shows up in thousands of independently built apps, it's a pattern, and it comes from the tool writing the code that none of its users read.
The researchers also point to a contradiction. Lovable advertises that you don't need technical knowledge to build an app. Yet its terms of service put responsibility for security and data protection on the customer. Non-experts, of all people, are the least likely to spot holes like these. Crissy Field therefore recommends having experts check access permissions before launching with real data.
Lovable's response and earlier incidents
According to Cybernews, Lovable doesn't see a vulnerability in its AI. It calls this a known, industry-wide configuration problem with Supabase. That's not entirely wrong, since the issue goes beyond Lovable. Security firm UpGuard examined around 300,000 domains connected to Supabase and found readable tables in 16,326 databases. More than half of them showed signs of personal data.
Lovable now offers a Security Scan and an AI security reviewer that check apps for these kinds of mistakes before they're published. Both arrived after a similar hole in Lovable apps went public in 2025 as CVE-2025-48757. CVE numbers are a globally used identifier for known security vulnerabilities. Startup Fortune points out that the new investigation covers apps built long after these tools launched. So the checks apparently miss too many mistakes. The magazine writes:
A warning that still lets 7,403 sites go live with open databases isn't protection. It's a disclaimer.
An industry-wide problem
Lovable is the star of the vibe coding wave. The Stockholm startup was most recently valued at $13.3 billion (about 12 billion euros) and hit $400 million (about 360 million euros) in annual revenue in the spring. Just recently, the company had to explain why it had promoted its AI with a hand-built designer website.
Its competitors face the same question. Bolt.new and Vercel's v0 also connect Supabase as their database. Replit and Cursor generate complete backends on request, meaning the invisible server side of an app with its database and logic. Supabase itself is riding the wave and raised $150 million (about 135 million euros) just a few weeks ago to get its database ready for AI agents.
Our take
The investigation highlights a familiar weakness of AI tools that build software on request: they deliver what you ask for, and that's usually an app that works. That doesn't automatically make it secure. If you don't know a database needs access rules, you won't notice when they're missing. The people who pay the price are mostly the users of these apps: customers who bought a ticket, ordered food or joined a forum without any idea how the site was built.
Who's to blame is up for debate. Lovable points to Supabase and to the responsibility of the people building the apps. Critics like Startup Fortune say the platform has to own this, because it writes the code and explicitly tells non-experts they don't need technical knowledge. It's still unclear how much of the data found is real and how much is just test data. Crissy Field itself stresses that the numbers aren't a confirmed leak rate. It's also unclear whether anyone unauthorized has already exploited the holes.
If you build apps yourself with Lovable, Bolt.new, v0 or similar tools, the takeaway is clear: before real customer data goes into your app, have someone with the right expertise check your database's access rules. Based on these findings, the built-in security scan alone isn't enough.
Sources
- Merkliste: trendingtopics.eu trendingtopics.eu
- sifted.eu sifted.eu
- startupfortune.com startupfortune.com
- wired.com wired.com
- dev.to #ai dev.to
- dev.to #ai dev.to
- dev.to #ai dev.to
- dev.to #ai dev.to
- dev.to #ai dev.to
- dev.to #ai dev.to
- dev.to #ai dev.to
- dev.to #ai dev.to
- dev.to #ai dev.to
Ask the AI assistant
Questions about this article, installation or background? Our assistant knows this page and runs on our own machine, not in someone else's cloud.
The Assistant is an AI (Gemma 4) and can make mistakes – please check important details against the source. Questions are used only to generate an answer and are deleted after one hour. Please don't enter any personal data.
Comments
What do you think? Already tried it? Write it here, no sign-up needed.
Comments are public. Links aren't allowed, and we delete abusive posts. You can delete your own comment at any time in this browser. Learn more in the Privacy Policy.