Personal Data Exposed in Lovable-Built Apps? Causes and What You Can Do
機械翻訳 / Machine-translated

機械翻訳 / Machine-translated
@aifriends
AI Friends(https://aifriends.jp)のクロスポスト公式アカウント。AIツールの紹介・使い方・できることを、中学生でもわかるやさしい日本語で届けます。
In an age where anyone can build an app by asking an AI, what would you do if your app was exposing other people's personal information for all to see? That is exactly the kind of problem that was discovered in apps built with Lovable. This article walks through what happened, why it happened, and what creators should do about it.
According to reporting by GIGAZINE, some apps built with Lovable — a service that lets you create web apps using AI — were displaying information even to users who were not logged in.
The types of information that were visible include:
The investigation was conducted by German security firm Crissy Field GmbH, in collaboration with German business newspaper Handelsblatt, which published the findings on November 2, 2026.
It has not been confirmed whether the exposed data was actually misused.
According to the Crissy Field report, the investigation covered 159,041 Lovable projects created between December 2025 and September 2026.
Of these, 34,545 used Supabase — a database service used to store app data.
Among those, 24,664 sites transmitted Supabase data at load time without requiring a login. Furthermore, 7,403 sites were found to have data containing email addresses.
That said, not all 24,664 sites represent a danger. Some data — like a news site's article listing — is meant to be publicly accessible by design.
The report itself notes that the 7,403 figure (approximately 30%) does not represent a definitive rate of data leakage, since email addresses may be public contact information or test data.
The investigation also used samples drawn from Common Crawl, a publicly available dataset of web pages, so the numbers do not represent all Lovable apps.
The core cause is insufficient RLS (Row-Level Security). RLS is a mechanism that determines, on a per-row basis in a database, who can read and who can write each record.
In a comment on Hacker News, Lovable's head of security explained that the affected apps had inadequate RLS configurations.
To make this more concrete: imagine a small restaurant owner who asked an AI to build a reservation app. The interface looks great and reservations come in without a hitch. But if the backend database has no "lock" specifying that only the person who made a reservation can view it, anyone who takes a closer look at the app's network traffic could read the names and phone numbers of every customer. The data isn't displayed on screen — but it's sitting wide open.
Crissy Field's researchers pointed out that Lovable's database integration follows the same pattern across all apps. This means a technique that works on one app could potentially be applied to others with relatively little effort.
Lovable's head of security stated that all apps are automatically scanned before publication. If a problem is found, users can select "Fix" to receive assistance with remediation.
Several future countermeasures were also outlined:
The head of security also stated that it is "not ideal for apps to still be leaking data."
On the other hand, the comment section saw significant criticism along the lines of "you're putting the responsibility on users." In particular, many took issue with the fact that security measures are optional in a service that markets itself to people without programming knowledge.
In fact, this is not the first time a problem of this kind has been found in Lovable. According to a June 2025 article on GIGAZINE, developer Matt Palmer reported a similar RLS misconfiguration (CVE-2025-48757).
At that time, 170 out of 1,645 web apps built with Lovable had the same issue. The first site discovered had the email addresses of roughly 500 people visible to anyone.
Comparing the two cases:
The larger scale of investigation naturally uncovered more. It is clear that the same type of misconfiguration has been repeated even a year later.
For an experienced engineer, checking database permission settings is one of the first things on the list. But with vibe coding — the practice of building an app by talking to an AI — it is easy to feel like the app is "done" once it works.
A polished appearance and actual security are two different things. This gap is precisely what makes problems like this one easy to fall into.
The cases in this investigation were primarily centered on German apps. However, the same type of misconfiguration could just as easily occur in Japan for apps built using the same underlying approach.
For example, imagine a non-engineer at a company building an internal inquiry management app using AI. The moment customer names and contact information are entered, it becomes a system that handles personal data. Even if the creator thinks of it as "just a small tool," the responsibility to protect that data remains unchanged.
Consider the perspective of a user, too. There may be situations where you send a passport or ID image to a small business's app. There is no way to tell from the outside whose hands built that app or what checks it went through.
As a user, the basic defense is to share only the information that is truly necessary.
Crissy Field recommends having a specialist review access rights, interfaces, and the separation of user accounts before handling any real customer data.
Even if consulting a specialist is not an option, there are minimum steps anyone can take:
Imagine an independent developer building a fan club app. If members' addresses or dates of birth are included, running through these five checks before launch can prevent a great deal of potential harm.
No. This investigation was based on a sample, and it cannot be said that all apps are at risk. However, unless creators verify settings such as RLS, the same problems can occur.
Crissy Field states that it was unable to confirm whether the data had been misused. However, it has also not been proven that misuse did not occur.
It is a mechanism that determines, on a per-row basis in a database, who is allowed to read or write each record. Think of it as putting a lock not just on the front door, but on each individual room.
Avoid carelessly sending important information — such as passport or ID details — to small or unfamiliar apps. Think carefully about whether the information is truly necessary before entering it.
The head of security has indicated plans to make publication blocking available across all plans and a goal of enabling automatic fixes by default. However, both are still in the planning or under-consideration stage.
If you are building an app with AI, the first thing to check before going public is: "Can data be seen when I'm logged out?"
This article is a cross-post from AI Friends.