An AI tool can turn a plain-English description into a working web app in an afternoon. A Hamburg security firm says that for a worrying number of those apps, the front door was left unlocked: one forum had more than 200,000 chat messages readable, and one service provider had passport numbers on show.
The research landed on 14 November on Hacker News, where it had passed 700 points by the morning of 15 November. Here is what it actually found, what it did not prove, and a lock-down checklist if you have built (or are about to build) an app with an AI builder.
What the researchers did
Crissy Field, a German application-security firm, worked with the business paper Handelsblatt, which reported the first findings on 2 November 2026. The firm’s own write-up says it used the open web archive Common Crawl to identify 159,041 projects built with Lovable, an AI app builder, published between December 2025 and September 2026.
They then focused on Supabase, a database service the researchers believe is the most widely used among Lovable projects. Of the 159,041 projects, 34,545 used it.
They looked at what each site handed over when it loaded, then made extra read-only requests with the same API key.
The numbers, read carefully
Headlines blur these figures, so here they are in order:
| Step | Count |
|---|---|
| Lovable projects examined | 159,041 |
| Of those, using Supabase | 34,545 |
| Sites that sent database records on load, with no login | 24,664 |
| Of those, sites with records containing email addresses | 7,403 (about 30%) |
Two cautions come straight from the report. Sending records on load “is not a security issue” by itself, because the data may be meant to be public. And the 30% figure “should therefore not be read as a confirmed data leak rate”, since emails can be public contact addresses or test data.
The sample is also not representative of all Lovable apps. The researchers could not say whether any exposed data was misused.
What was sitting in the open
The report lists specific cases, each from a single operator rather than the whole pool:
- A German lifestyle forum exposed more than 80,000 profiles (names, emails, birth dates, gender) and more than 200,000 chat messages.
- Event organizers and restaurants exposed tens of thousands of ticket and meal-sales records, including phone numbers, addresses, payment information and ticket codes.
- One service provider exposed more than 1,000 records with identity-document details: passport numbers, expiry dates and nationalities.
- Elsewhere: contracts, invoices, photo albums, internal WhatsApp communications, and 280,000 emails sent to one provider’s customers.
The researchers retrieved only parts of the datasets, so real totals could be larger, and they did not publish how to reach the data.
Why it happens
The report blames access-control mistakes and notes they “can happen in any application, whether written by people or built with AI”. Techzine’s coverage describes the cause as poorly configured Supabase databases.
The twist is repetition. Lovable’s database integrations follow recurring patterns, so a misconfiguration that works against one app may work against many.
The report also points to a gap between Lovable’s marketing, which stresses that no technical knowledge is needed, and its Terms & Conditions, which put responsibility for the app and its data on customers. According to the report, Handelsblatt contacted Lovable with the findings. None of the pages used for this article describes a reply, so check Lovable’s own channels for a response.
A lock-down checklist for AI-built apps
Supabase’s documentation describes Row Level Security (RLS) as “granular authorization rules that run inside the database”. It also warns that a table in an exposed schema without RLS “is readable and writable by any role with a grant on it”.
These steps come from Supabase’s and Lovable’s own docs. Run SQL only on a project you own, ideally on a test copy first.
- Turn on RLS for every table that holds user data:
alter table public.reports enable row level security;
With RLS on and no policies, Supabase says no data is reachable through the API with a publishable key.
- Trim the grants. Supabase notes that on existing projects, new tables in
publicstart with all privileges granted toanon,authenticatedandservice_role. Its example:
revoke all on table public.reports from anon, authenticated;
grant select, insert, update, delete on table public.reports to authenticated;
- Write per-user policies. Supabase’s example lets a signed-in user read only their own profile:
create policy "Users can view their own profile."
on profiles for select
to authenticated
using ( (select auth.uid()) = user_id );
Avoid to anon using ( true ) unless the data is meant to be public.
- Keep secret keys out of the browser. Supabase says: “Never use a secret key in the browser or expose it to customers.” Lovable’s docs add that you should not hardcode keys in frontend code; store them in Secrets and call external APIs from server-side code.
- Run Lovable’s scans. Per Lovable’s docs, the Quick security scan runs on every publish and flags tables without row-level security and rules that let everyone through. The Deep security scan also reviews application code. Policies can be reviewed under More, then Database, then Manage, then Policies. The docs do not say whether these scans cover a separately connected Supabase project, so confirm RLS in Supabase itself.
- Know the plan limits. Lovable’s sensitive data scanning, which can spot passports, needs an Enterprise plan and does not scan external or self-hosted resources.
- Re-check after big changes, and get a human review before real customer data goes in. The report says automated checks “are no substitute for a professional security assessment”. Lovable’s docs likewise say its tools cannot guarantee complete security.
If you used a small app that might be affected
The report names no affected apps, so you cannot tell from the research whether a given site was exposed. It does say exposed data of this kind could feed phishing, ticket fraud or identity theft. If a company tells you your details were exposed, our guide on what to do in the first 24 hours after a data breach notice is the place to start, and finding every account that uses your username helps you see where else your details may be sitting.
What happens next
As of 15 November 2026, the research is days old. The open questions are how many affected apps have been fixed and whether Lovable will respond. The study did not examine Lovable’s own newer database features, so those remain untested.
Sources
Links last checked on 15 November 2026.





Write a response