DEV Community

Cover image for 24,664 Lovable Apps Were Readable Without a Login. Here's the 10-Minute Check for Yours
Nirali for Toolbit AI

Posted on AI-assisted

24,664 Lovable Apps Were Readable Without a Login. Here's the 10-Minute Check for Yours

On 14 November a report from Hamburg security firm Crissy Field drew nearly 700 points on Hacker News. Working with the German business paper Handelsblatt, which first reported the findings on 2 November, they looked at Lovable apps archived by Common Crawl and found 24,664 sites that sent Supabase database records to the browser as soon as the page loaded, with no login.

Nobody hacked Lovable or Supabase. Lovable's Head of Security said it plainly in the HN thread: "The root cause in these apps was insufficent Row-Level Security."

If you've shipped anything with Lovable or any AI tool that wires up Supabase for you, the useful question isn't "how did this happen to them?" It's "is mine one of them?" This post is the check.

Funnel: 159,041 Lovable projects analyzed, 34,545 on Supabase, 24,664 returned records without login, 7,403 had readable tables with email addresses
Figures from Crissy Field's English report (November 2026).

What the report found

Step Count
Lovable projects found in Common Crawl (Dec 2025 to Sep 2026) 159,041
Using Supabase as the backend 34,545
Sent database records on page load, no login 24,664
Had readable tables containing email addresses 7,403

The data included more than 1,000 passport records from one service provider and more than 200,000 chat messages from a lifestyle forum. Event organisers and restaurants exposed ticket and food orders with phone numbers and addresses.

Two fair caveats from the researchers themselves: the 30% share (7,403 of the 24,664 sites) "should therefore not be read as a confirmed data leak rate", because an email address can be a public contact or test data, and they couldn't tell whether anyone had misused the data.

Why this keeps happening

Supabase apps talk to the database straight from the browser. To do that, the page carries a public key. That's by design: Supabase's docs say the publishable key is "Safe to expose online."

What protects your data is Row Level Security (RLS), a set of rules in Postgres that decides which rows each user can see. Without it, the public key is enough. The Supabase docs put it bluntly:

"A table in an exposed schema without RLS is readable and writable by any role with a grant on it."

AI builders make this easy to miss. You describe a feature, the tool creates a table, the app works, and nothing tells you that a logged-out stranger can read the same table.

The part most posts skip: the new fixes don't cover your old tables

Both companies have shipped changes. They're good changes. They just don't reach the apps that are already live.

Fix When What it covers What it doesn't
Supabase stops auto-exposing new public tables to the Data API Applied to all existing projects on 30 Oct 2026 Tables created after that Existing tables "keep their current grants and stay reachable"
RLS on by default Tables made in the Supabase dashboard Dashboard-created tables Tables made in the SQL Editor or "through another tool", such as migrations or an AI builder
Lovable asks you to review critical findings before publishing 12 Nov 2026 Projects on Lovable's built-in backend (not the Supabase-backed apps in this study) You can still choose "publish anyway"

So if your app went live before November, assume nothing has changed for it until you check.

The 10-minute check

Run these on your own project only.

1. Open Security Advisor. In the Supabase dashboard, go to Advisors, then Security Advisor. Any "RLS Disabled in Public" error is a table anyone can query. Fix those first.

2. List your tables without RLS. In the SQL Editor:

select schemaname, tablename, rowsecurity
from pg_tables
where schemaname = 'public'
order by rowsecurity, tablename;
Enter fullscreen mode Exit fullscreen mode

Anything with rowsecurity = false needs attention.

3. Try it logged out. Use your project URL and the same public key your frontend uses, against a table you know exists:

curl "https://<project_ref>.supabase.co/rest/v1/<table>?select=*&limit=5" \
  -H "apikey: <publishable_or_anon_key>"
Enter fullscreen mode Exit fullscreen mode

If a private table returns rows, so can anyone else's request. You can also open your live app in a private window and watch the /rest/v1/ calls in DevTools.

4. Turn on RLS and add an owner-only rule. For a table where each row belongs to a user:

alter table public.your_table enable row level security;

create policy "Users can read their own rows"
on public.your_table for select to authenticated
using ((select auth.uid()) = user_id);
Enter fullscreen mode Exit fullscreen mode

Add matching policies for insert, update and delete. Avoid using (true) on anything sensitive. It's the same as having no rule.

5. Check storage buckets. Uploaded files are a separate door:

select id, public from storage.buckets order by public desc, id;
Enter fullscreen mode Exit fullscreen mode

Supabase says a public bucket "effectively bypasses access controls". Passports, invoices and contracts belong in private buckets, served with signed URLs.

6. Never ship the secret key. The service role key ignores RLS completely. If it's anywhere in your frontend code, rotate it now.

If you built with Lovable, also run its Deep scan (free to run; asking for it in chat uses credits). Lovable's own FAQ says the scans "do not guarantee" security, so treat them as a first pass, not a sign-off.

This isn't a one-off

  • CVE-2025-48757, published in May 2025 after Matt Palmer reported it to Lovable in March, covered 303 endpoints across 170 Lovable projects (about 10.3% of 1,645 checked) with inadequate RLS.
  • In September 2026, UpGuard reported 16,326 Supabase databases exposing readable tables, out of roughly 300,000 domains showing signs of Supabase use.

Different researchers, same root cause. AI tools write the app quickly, and the access rules are the step a human has to check.

If you're comparing scanners to add to that checklist, Toolbit keeps a list of AI security tools you can look through.

Have you run the logged-out test on something you shipped with an AI builder? I'm curious how many people find an open table on the first try.


This article was created with the help of AI and reviewed by the author for accuracy. Disclosure: I work at Toolbit, linked above.

Sources

Top comments (0)