Earlier this year, researchers scanned 159,041 projects published on Lovable — an AI-powered “vibe coding” platform that lets anyone describe an app in plain language and have it built automatically. What they found was not an edge case. Of the 24,664 projects examined closely, 7,403 were exposing personal data without requiring any login to view it. The data included passport numbers, payment records, and personally identifiable information that people had entered into these apps in good faith.
This is the vibe-coding privacy crisis: a category of data exposure that is growing because the tools making software development accessible to non-engineers are, in many implementations, not teaching those engineers about the security configurations their apps require.
What Vibe Coding Is
Vibe coding refers to building software by describing what you want in natural language and having an AI generate the code. Tools like Lovable, Bolt, and others let someone with no programming experience create functional web applications — a customer scheduling form, a small-business invoice tracker, a club membership registry — in minutes.
The democratization is real. These tools have meaningfully lowered the barrier to creating software. The problem is that building software that works is not the same as building software that handles personal data responsibly.
A traditional software developer learns, over years of training or experience, that databases containing personal data need access controls. They know what “row-level security” means. They know that exposing a database API key on the client side — in code that runs in someone’s browser — allows anyone with that key to query the database directly.
A person building an app via AI prompts typically doesn’t know any of this. They’ve been told that the tool will handle the technical details. And for most technical details — building a form, creating a user interface, connecting to a database — the AI does handle them. Access control is the category where the gap becomes visible.
Why Supabase Gets Misconfigured
Most vibe-coded apps that store data use Supabase as the backend database. Supabase is a good tool for rapid development. It provides a Postgres database, an authentication system, and a set of APIs that allow the front end to query data. The problem lies in how access control works by default.
Supabase provides every project with a public API key — a credential that allows the front end to connect to the database. This key is intended to be used in combination with Row Level Security (RLS): database policies that restrict which rows each user can read or write, based on their authenticated identity.
Without RLS enabled, the public key allows any caller to retrieve any data in any table. If an AI coding tool generates an app that uses the public key but doesn’t configure RLS policies, the database is effectively public. Anyone who knows the API key — and the key is typically visible in the app’s front-end code, which runs in the browser — can query it.
In September 2026, TechCrunch reported on research by UpGuard that found 16,326 Supabase databases with publicly readable tables exposed to the internet. Over half of them appeared to be leaking personally identifiable information. This was not a Supabase security vulnerability. Supabase’s documentation explicitly recommends enabling RLS and warns about the risks of not doing so. The misconfiguration happened at the application layer, by developers — many of them non-technical — who did not know to configure it.
What Data Gets Exposed
The nature of the exposed data follows the nature of the apps. Vibe-coded apps are often built to solve specific personal or small-business problems. That means the data inside them tends to be:
Personally identifiable: Names, email addresses, phone numbers, home addresses. These are the minimum required fields for most useful applications.
Sensitive documents: One category found in the Crissy Field research scan of Lovable projects was passport numbers. Apps built for travel agencies, visa processing, identity verification, or document management frequently ask users to enter or upload government IDs. Without access controls, that data is readable by anyone.
Financial records: Invoice trackers, expense managers, and payment processors that store card numbers, bank account references, or transaction histories. The financial data category is particularly valuable for downstream fraud.
Health and personal information: Apps built for personal health tracking, medication management, or therapy scheduling often store sensitive medical context that users entered expecting it to remain private.
The Lovable Platform Specifically
Lovable had a compound problem. Beyond the Supabase misconfiguration issue common to many AI coding platforms, a separate April 2026 report documented a breach in which source code, Supabase credentials, and chat transcripts were exposed for projects created before November 2025. That exposure meant that even projects that had properly configured their databases might have had their credentials compromised, allowing a different kind of access.
The researchers who scanned 159,041 Lovable projects found that 7,403 — roughly 4.6% of the total — were actively exposing personal data without authentication. This isn’t a tail-risk scenario in a few hundred edge cases. It’s a systematic misconfiguration at a scale that reflects how many apps are being built without anyone verifying that the data they store is actually protected.
Lovable has updated its platform since this research became public, adding warnings and guardrails around Supabase configuration. The existing apps, however, aren’t automatically remediated.
How to Check if an App You’re Using Was Built This Way
There are observable signals that an application you’re using might have this class of problem.
Check the technology stack. If you can see “supabase.co” appearing in browser network requests while using an app, the app uses Supabase. This isn’t by itself a problem — Supabase is a legitimate and well-regarded service. But it means the security of that app’s data depends on whether the developer properly configured access controls.
Look for RLS warnings in public changelogs. If the app has a public changelog or git history (many AI-built apps publish to GitHub), search for references to “RLS” or “row level security.” A recent commit adding RLS policies is a signal that it wasn’t there before.
Consider the developer’s background. If the app appears to be a solo project or a small business tool, and there’s no information suggesting the person who built it has a software engineering background, treat any personal data you enter with proportional caution. This is not a judgment — the vibe coding tools often produce excellent, functional software — it’s a risk calibration.
Use disposable or minimal information when possible. For apps built by individuals or small teams where you can’t verify the security posture, use a secondary email address, avoid uploading sensitive documents, and prefer entering minimal information rather than complete profiles.
What This Means for Where You Store Sensitive Personal Data
The vibe-coding privacy crisis is a specific manifestation of a broader principle: the infrastructure behind an app determines what actually happens to your data, and the app’s surface-level design tells you almost nothing about whether that infrastructure is secure.
An app can look polished, work reliably, and collect payment while having its database open to anyone who hits the right API endpoint. The user experience layer and the data security layer are independent concerns, and the former has been almost entirely automated by AI coding tools while the latter has not.
This is a reason to be more selective about which apps you trust with sensitive personal information — specifically photos, identity documents, medical records, and anything else that would cause real harm if it became public. The question is not whether the app looks trustworthy, but whether the people who built it have the knowledge or the resources to have configured its infrastructure securely.
For private personal file storage in particular, purpose-built infrastructure with explicit security review is a different category than a vibe-coded app assembled with a prompt. daftei stores files under AES-256 encryption at rest and TLS 1.3 in transit, operates under GDPR and CCPA compliance frameworks, and never sells data or trains third-party AI on user content. The security properties are stated explicitly and are auditable, rather than being a function of whether a developer configured a checkbox during code generation.
The Broader Pattern
The vibe-coding privacy crisis is not, fundamentally, about Lovable or Supabase. It’s about a gap between capability and responsibility that emerges when powerful tools are made accessible before the people using them have the context to use them safely.
A word processor can produce a legal document. That doesn’t mean the person who wrote it has the legal knowledge to know if it’s enforceable. An AI coding tool can produce an app that handles personal data. That doesn’t mean the person who built it knows how to configure a secure backend.
The gap closes over time, as platforms add better defaults, as users gain experience, and as the egregious examples become case studies. In the meantime, the practical implication is to treat any app handling your personal data as being only as secure as you can verify it to be — and to be more cautious than the app’s design might suggest.