Search TorWire

Find cybersecurity guides and research articles

Home > News > Cybersecurity > Lovable Sites Leak Private Messages, Passport Data and Payment Information

Lovable Sites Leak Private Messages, Passport Data and Payment Information

By: Morgan Cipher — Senior Privacy Journalist

Last updated: November 3, 2026

Human Written
Lovable Sites Leak Private Messages, Passport Data and Payment Information
  • Researchers found 7,403 Lovable-built sites with readable email data. Some also exposed passport numbers, payment details and private chats.

  • A missing database lock causes the problem. AI-written apps often skip it, and the app preview never shows the gap.

  • Lovable’s terms make customers responsible for their apps and data. The company told Handelsblatt the problem exists across the industry.

Apps built with the AI tool Lovable left private data open to anyone online. Visitors needed no login.

German security firm Crissy Field teamed up with the business daily Handelsblatt, which reported the story on November 2.

The open data had passport numbers, payment info, and over 200,000 private chat messages out in the open. 

How the Researchers Found the Unprotected Database

Crissy Field started with Common Crawl, a public web archive. From it, the team found 159,041 Lovable projects crawled between December 2025 and September 2026. Of those, 34,545 used a database service called Supabase. Next, the team opened the sites in a browser without logging in. In all, 24,664 sites loaded database records on their own.

Then the team used each site’s own public key to run read-only checks. As a result, it found email addresses in 10,560 tables across 7,403 sites. That equals 30% of the 24,664 sites.

However, the report warns against treating 30% as a leak rate for all Lovable apps. Some emails may be public support addresses or fake test data. Also, the archive is not a random sample of Lovable’s projects.

Still, the real numbers probably run higher. Researchers pulled only small slices of each table. They also kept site addresses and attack scripts private.

What Anyone could Read

The worst cases involved personal and money data.

  • A single service provider revealed over 1,000 records containing passport numbers, expiry dates and nationality.
  • The German lifestyle forum also published more than 80,000 user profiles, including names, email addresses, birth dates and sex. Furthermore, over 200,000 confidential messages were exposed.
  • Event hosts and restaurants exposed tens of thousands of orders. Those records held payment details and working ticket codes.
  • A property firm left AI-written summaries of its internal emails open.
  • Anyone could read 280,000 emails that one provider had sent to its customers.

Passport data matters most. A bank can cancel a stolen card. But a passport number stays the same for years. Paired with a name and nationality, it helps criminals commit fraud.

Why the Lock was Missing

Most of Lovable’s data-focused projects run on Supabase. The app code isn’t stored on a server it runs right in a visitor’s browser and communicates directly to the database. Furthermore, the key that makes this connection is public by design. So the database itself must decide who sees what.

A feature called Row Level Security does that job. When it is off, anyone with the public key can read the table. Supabase’s own guide says so.

One analysis of the findings says Lovable’s AI agent often skips this step. The agent writes raw database code, and tables made that way start unlocked. Turning the lock on without the right rules breaks the app during testing. So the fastest path for the AI is to leave the table open.

Autonomous AI agents can also pose risks beyond exposing data. In another incident, an AI agent deployed custom ransomware targeting AI models, highlighting how automated systems can enable destructive cyberattacks.

Nothing warns the builder, either. A locked database and an open one look the same in the preview. Meanwhile, Lovable’s AI builds apps with similar table names, such as “messages” and “orders.” That lets an attacker run one script against thousands of sites.

Who is Responsible for Securing App Data

Lovable promises that users need no coding skills. Its terms, however, make customers responsible for their apps and data. Even so, the company says it has hosted more than 50 million projects.

Lovable told Handelsblatt the problem affects the whole industry. Since the 2025 reports, the company has added a security scan and an AI reviewer. Yet the analysis argues these tools cannot tell which tables hold secrets. It urges locked tables by default and human reviews before real customer data goes in.

Handelsblatt adds that its tip came in mid-August. It writes that thieves may have had access for at least a year. The paper also says tens of thousands of sites appear affected. That claim is wider than the 7,403 sites above.

Not Lovable’s First Warning

Previous incidents provide evidence in support of this assertion. Initially, in 2025, a researcher examined 1645 applications within Lovable’s public directory. The researcher discovered 170 applications with information displayed in tables that were tracked as CVE 2025 48757. Lovable disagreed with this finding and insisted that the responsibility of protecting their data lies with clients.

Later, in April 2026, a security researcher found another flaw that made it possible for any free account holder to get access to the code and AI chats of other users related to the older projects.

Lovable initially claimed that it did not suffer from a security breach, but later on it confessed that its explanation of the meaning of the term ‘public’ was ambiguous. In that case, it was the visibility of the project that was in place, but not the database of the clients.

The app owner should activate Row Level Security for every file. After that, they should analyze what a person who is not logged into the system can gain access to. Only after conducting an analysis, they may proceed with storing information about a passport or other payment details, and so on.

Customers cannot check a database themselves. Still, they can limit the harm. For example, share only what a service truly needs.

Experts also recommend rotating API keys from public projects. This entails changing old keys to new ones, so in case the old ones get compromised, whoever has them would lose access. Also, watch for phishing messages that use your real details.

Share this article

You might also like

Hackers Claim Theft of 436,000 Patient Records From French Pharmacies

Hackers Claim Theft of 436,000 Patient Records From French Pharmacies

Hackers say they stole records on over 436,000 people from French pharmacies in Paris, Strasbourg, and Dijon. Nobody has proven…

November 3, 2026
FBI Director Says AI Use Jumped 605 to Stop Threats

FBI Expands AI Use by 605% to Speed Threat Detection, Director Says

The FBI Director, Kash Patel, says AI usage at the bureau has increased by 605 percent, but the precise details…

November 2, 2026
Russia-Linked Hackers Target Industrial Systems Across Spain, Denmark, and the US

Russia-Linked Hackers Cause Physical Disruption at Industrial Sites in Europe and the US

Hackers linked to Russia reportedly broke into weakly protected factory, water and food systems across Spain, Denmark and the United…

November 2, 2026

About the Author

Morgan Cipher

Morgan Cipher

Senior Privacy Journalist

Morgan combines a journalist’s curiosity with a security specialist’s precision. His reporting on data breaches, privacy laws, and encryption tech has been featured in several tech publications. At TorWire, he focuses on real-world threats and how to counter them, always with an eye on what’s next in digital privacy.

Comments (0)

No comments.