Skip to content
Developer312

Lovable Database Research Makes Authorization the Launch Gate

Updated research on Lovable apps finds sensitive Supabase records accessible without login. The figures demand permission tests, not claims that 30% of all apps leaked.

Published Report an error

Key Takeaways: What Matters Most?

  • Crissy Field analyzed 159,041 projects, but its Common Crawl sample is not representative of all Lovable applications.
  • The 30% email-record finding is not a confirmed breach rate and applies only to 24,664 more closely examined websites.
  • Researchers describe passport details, private messages and internal email summaries accessible without login.
  • Public browser keys are not substitutes for database authorization rules and appropriately limited grants.
  • Builders should test signed-out access and cross-account isolation before introducing customer data.

Takeaways like these, every weekday morning. Get the briefing

What Actually Happened

Updated research from Crissy Field, covered by StartupHub on November 15, describes sensitive information accessible without login in applications built using Lovable and Supabase. The operational signal is specific: generating an application with authentication screens does not establish that its database enforces authorization.

The research is not a newly discovered platform-wide breach dated November 15. Crissy Field says it investigated with Handelsblatt, whose initial article appeared November 2, and that its current report provides newer analysis and updated figures. The exact publication time of that update is not established in the retrieved report. Monday's reason to examine it is the weekend coverage and the concrete access-control evidence, not an invented fresh incident date.

Using Common Crawl datasets, the firm identified 159,041 Lovable projects from December 2025 through September 2026. Of those, 34,545 used Supabase. It found 24,664 websites returning database records on initial loading without login. That behavior alone is not a vulnerability: many websites intentionally publish database-backed content.

Additional read-only requests using the same API key retrieved records containing email addresses in 10,560 tables across 7,403 of those websites. The researchers explicitly caution that the resulting 30% figure is not a confirmed data-leak rate. Some addresses may be public or synthetic; sensitive records without email addresses escape that measurement. The sample is also not representative of all Lovable apps.

The serious evidence is in the records, not the headline percentage

Crissy Field describes a German lifestyle forum exposing more than 80,000 profiles and more than 200,000 chat messages. A separate service provider exposed over 1,000 identity-document records containing passport numbers, expiration dates and nationalities. Other examples included ticket-sale information, internal communications, contracts and AI-generated summaries of property-management emails.

Those are researcher-reported observations, not independently reproduced findings by Developer312. StartupHub's reporting provides a separate account of the research, not a second technical scan. Neither source establishes that criminals subsequently used the exposed material. Crissy Field explicitly says it could not determine misuse.

That limitation does not make unauthorized availability harmless. A database can create a confidentiality incident before anyone can prove fraud, and summaries generated from private emails can retain sensitive facts even when the original messages are not exposed. Teams should classify derived records according to their contents, not assume an AI-produced summary is safe to publish.

A login screen is not a database policy

Supabase's documentation separates grants, which determine whether a database role may perform an operation, from row-level policies, which determine the rows it may affect. Both belong in the review. A frontend that hides a page does not prevent direct requests to an API that permits access.

The correct response is not to treat every browser-visible key as a leaked administrative credential. Client applications need a way to reach their backend. The important question is what permissions a request receives and which rows it may read or modify. Administrative service-role access, which bypasses row-level security, requires a different server-side boundary.

This is where AI-assisted development changes the operating economics. Reusable integration patterns make implementation faster, but repeating an unsafe pattern can repeat its exposure. A successful generated demo provides no evidence that signed-out visitors, ordinary users and administrators receive different permissions at the database layer.

For a small company, an authorization review is therefore part of the cost of accepting customer data. It should not be deferred until the first enterprise questionnaire. The research supports paying attention to permissions; it does not quantify the cost of a review or establish that a specific scanner would have prevented every case.

What Builders Should Take From It

  • Create an access matrix before launch. List which roles can read, create, change and delete each sensitive record type. Explicitly distinguish public content from account-specific information.
  • Test denied access as carefully as allowed access. Use synthetic records in an environment you own. Confirm that signed-out requests fail where intended and that one ordinary account cannot retrieve another account's private rows.
  • Include derived data. Review summaries, support messages, exports and internal dashboards alongside the original customer tables. Convenience copies can carry the same confidentiality obligations.
  • Repeat the checks after schema changes. New tables, views and generated features can add access paths even when the original release passed its tests. Keep the permission change and its tests together.

If an exposure is suspected, preserve the relevant configuration and access records, restrict unnecessary access through an authorized response, and determine the affected data before making notification claims. Do not explore other people's exposed databases to validate a headline.

The bottom line is not that AI app builders are unusable. It is that their speed moves the security decision earlier: an app that will hold confidential data needs proven authorization before the first real record, not after the first impressive demo.

Disclosure: This article was prepared with AI assistance from the cited sources. It has not received human editorial review.

Developer312 covers the AI business signals builders actually need to act on. Get the weekday briefing at developer312.com.

Sources

  1. [1]Lovable: Many Apps Built on the Platform Expose Sensitive Data (opens in a new tab) — Crissy Field
  2. [2]Lovable apps left passports and chats exposed (opens in a new tab) — StartupHub.ai (2026-11-15)
  3. [3]Row Level Security (opens in a new tab) — Supabase

Frequently Asked Questions

Get the next briefing

The lead story each weekday morning, with the key takeaways and a link to the full analysis.

One lead story each weekday morning: what happened, the key takeaways, and a link to the full analysis.

New posts daily, briefing weekday mornings. We use your email only to send the briefing. Unsubscribe anytime. Privacy policy.

Related Articles