159,041 projects. 34,545 of them on Supabase. 7,403 whose databases gave up records containing email addresses to read-only queries made with the same key the site itself loads. That is Crissy Field's survey of Lovable apps, done with Handelsblatt and near the top of Hacker News on Saturday. Among what came back without a login: 80,000 forum profiles and more than 200,000 private chat messages, ticket and payment records from event organizers and restaurants, and more than 1,000 records with passport numbers, expiry dates and nationalities.
The authors are careful that the 30% is not a confirmed leak rate; some of those addresses are public contacts or test data. The passports need no such caveat. But the line that should worry anyone building on a platform like this sits in the report's last section, not its first:
With Lovable, database integrations follow recurring patterns, making it easier for attackers to look for similar misconfigurations across many applications. An approach that works against one app may then work against others with little extra effort.Crissy Field
Hand-written software fails idiosyncratically. Each team makes its own mistakes, and an attacker finds them one codebase at a time. Generated software fails in the shape of its generator. That is the bargain Lovable sells: every app gets the same plumbing, so nobody has to understand the plumbing. Its marketing says no technical knowledge is needed; its terms, the report notes, put responsibility for the app and its data on the customer. Both are true at once. The passport numbers sat in the gap.
Generated software fails in the shape of its generator.
That is why the researchers' prescription is the wrong size. Crissy Field recommends an expert review of access permissions before an app handles confidential data, repeated after significant changes, with outside help when the expertise isn't in-house. Sound advice for a codebase. Here it means tens of thousands of separate reviews, bought by people who picked Lovable so they wouldn't have to hire anyone who knows what an access policy is. (Crissy Field, for the record, sells that review.)
The strongest objection is that only the owner can do this work. A platform can't know which tables are meant to be public. A restaurant's menu should be readable by anyone; its orders shouldn't. Access control is the owner's job on any stack, and a developer writing every line by hand can leave a table open just as easily. The report says as much.
True, and beside the point. In an app built from a plain-language description, the owner was never asked which tables were public. The access rules came from the same place as the rest of the app: the generator. Whatever Lovable checks before an app goes live, the passport records are the measure of how well it works. And the sameness that let one technique run across thousands of apps runs the other way too. A default tightened once, or a pre-publish check that shows the owner in plain words what a stranger holding the page's key can read, reaches every app on the platform at the same moment. An attacker gets that leverage for free. A defender gets it only by owning the template.
Built alike, the apps broke alike. They can be fixed alike, too, but only by the party that wrote the template.