Murat BekgiPenetration tester
Write-ups

Week 47: 7,403 Lovable apps leaking data, and how to test yours

Researchers read private data from 7,403 Lovable apps without logging in. Here's the exact test, step by step, plus Java4Shell and the Optimizely skimmer.

Published
Reading time
3 min
Topic
Weekly
Contents
  1. What people were reading
  2. Walkthrough: reading a Lovable app’s database without logging in
  3. Also patch this week

November 11 to 17. The most-read security stories this week, then a hands-on walkthrough of the one you can actually test yourself.

What people were reading

These are the security and tech stories that got the most upvotes on Hacker News this week. The number is how many people upvoted each one. More upvotes, more people talking about it.

Not security, but people were talking about them: SpaceX caught Starship after Flight 15, OpenAI released GPT-6.2 and Rust 1.100 shipped.

Walkthrough: reading a Lovable app’s database without logging in

Crissy Field scanned 159,041 apps built with Lovable, the AI app builder. On 7,403 of them they could read private data with no login at all: passport numbers, payment records, and over 200,000 chat messages from one forum (Crissy Field, Techzine).

The bug is simple, and it shows up again and again in AI-built apps. Lovable apps store data in Supabase. The browser talks to Supabase directly with a public key, which is fine and by design. What protects the data is row-level security (RLS): rules that say who can read which rows. If nobody turns RLS on for a table, the public key can read the whole table.

Here’s how to test it. Only do this on an app you own or have written permission to test.

1. Build a target. Create a free Supabase project. Add a table called profiles with a few fake rows (name, email, phone). Don’t enable RLS. That’s the mistake you’re reproducing.

2. Find the URL and key, the way an attacker would. Open your app in the browser, press F12 and go to the Network tab. Reload and click any request to *.supabase.co. Two headers matter: the project URL, and the apikey header. That key is the public “anon” key. It ships in every Supabase front end, so anyone can grab it.

3. Ask for the table. Supabase puts a REST API in front of every table. Replace the URL and key with yours:

curl "https://YOUR-PROJECT.supabase.co/rest/v1/profiles?select=*" \
  -H "apikey: YOUR-ANON-KEY" \
  -H "Authorization: Bearer YOUR-ANON-KEY"

With RLS off, you get every row back as JSON. That’s the whole attack. No password, no exploit code.

4. Guess other tables. Attackers don’t stop at one. Try names the app probably uses: users, orders, messages, payments. A table that doesn’t exist returns an error. One that exists without RLS returns data.

5. Fix it. In the Supabase SQL editor, turn RLS on and allow users to read only their own row:

alter table public.profiles enable row level security;

create policy "read own profile"
on public.profiles for select
using (auth.uid() = id);

6. Test again. Run the same curl. You should get back [], an empty list. The key still works, but RLS now stops it from reading anyone’s row.

To find every unprotected table in a project you own:

select tablename
from pg_tables
where schemaname = 'public' and rowsecurity = false;

Every row this returns needs RLS and a policy before it holds real data.

Also patch this week

Java4Shell (CVE-2026-114938). A memory bug in Java’s Base64 decoder lets an attacker run code on the server without logging in (Calif). It hits OpenJDK 17.0.6 and newer, including Amazon Corretto, Temurin and Red Hat’s builds, but only on Intel and AMD chips with a feature called AVX-512 VBMI. ARM servers like AWS Graviton aren’t affected (Unite.AI). Fixed versions came out November 17 (Upwind). On a Linux server, this tells you whether you’re exposed:

java -version 2>&1 | head -1
grep -o -m1 avx512vbmi /proc/cpuinfo || echo "this CPU can't trigger Java4Shell"

If the Java version is 17.0.6 or newer and avx512vbmi prints, update Java. Updating your app’s libraries won’t fix it, because the bug is in Java itself.

Optimizely skimmer. On November 16, scripts served from Optimizely’s CDN started loading a card skimmer on pages at CNN, BBC, Zoom, Salesforce, GitLab and KLM (StepSecurity). Report URI counted 178 affected organizations (Report URI). If you test checkout pages, list every outside script that loads there. Each one can read the card form.

Want this every week? Subscribe.