Short answer: in November 2026 a Hamburg security firm, Crissy Field GmbH, reported that many apps built on Lovable expose personal data without a login. Lovable's head of security has said the root cause in these apps was insufficient Row-Level Security (RLS). If your app uses Supabase, you can check your own exposure in about ten minutes: ask the database API for a whole table using the public anon key, and again using another user's token. If any row comes back that should not, RLS is off or a policy is wrong, and that is the thing to fix.
This article summarises what was published, with the researchers' own caveats, and then gives the test. It does not claim that any specific app is affected, and it is not legal or security advice.
What was found
Crissy Field GmbH published its research on its own site, Lovable: Many Apps Built on the Platform Expose Sensitive Data, and worked with the newspaper Handelsblatt, which reported on 2 November 2026 (Lina Knees). The figures from the researchers' page:
| Step | Number |
|---|---|
| Lovable projects identified from Common Crawl (December 2025 to September 2026) | 159,041 |
| Of those, projects that used Supabase | 34,545 |
| Sites that sent Supabase records as soon as the page loaded, with no login | 24,664 |
| Sites where further read-only requests returned tables with email addresses | 7,403 (30% of the 24,664) |
| Tables with email addresses across those sites | 10,560 |
The researchers are explicit about what this does not mean. Email addresses can be public contact addresses or test data, and the approach misses personal records without an email address, so the 30% should not be read as a confirmed leak rate. It applies only to the 24,664 sites examined more closely, and the Common Crawl sample is not representative of all Lovable apps. Sending records on page load is also not a problem by itself, because the data may be meant to be public.
Examples they list from the records they actually downloaded:
- A German lifestyle forum exposed more than 80,000 profiles (names, email addresses, dates of birth, gender) and more than 200,000 chat messages.
- Event organisers and restaurants exposed tens of thousands of ticket and food-sales records, including email addresses, phone numbers, postal addresses, payment information and ticket codes.
- A property management company exposed AI-generated summaries of internal emails about tenants, owners and business partners.
- A service provider exposed more than 1,000 records with identity document details.
The researchers did not publish how to access the data, and say they could not establish whether any of it was misused. They also note that their report presents a more recent analysis, with updated figures, than the Handelsblatt article.
What Lovable has said
The story reached a wide audience on Hacker News on 14 November 2026. In that thread, Lovable's head of security, commenting on Hacker News, said the root cause in these apps was insufficient Row-Level Security, which the platform's pre-publish scan is meant to catch, and that a free deeper scan is available. I could not verify the details of that scan, so I leave it to Lovable's own documentation. The point for you is narrower: the exposure mechanism named on all sides is the database access rules, not a mysterious flaw you cannot see.
Why this happens
Supabase exposes your tables through an HTTP API. The browser needs a public key (the anon key) to talk to it, so that key is visible to anyone who opens your site, by design. What stops a stranger from using it to read everything is not the key, it is RLS: rules on each table that say which rows a given user may read or change. A Supabase community discussion from 2021 shows the same surprise from the other side: a developer who used Supabase only for login realised that the key needed in the frontend also grants database access by default, and asked how to switch that off for every table at once.
AI-generated apps make this easy to miss, for a reason that has nothing to do with intent. The screens usually ask for rows with a filter such as "user_id equals the signed-in user". User B only ever requests B's rows, so the app looks correct. But the database was never told to refuse anything. Anyone who skips the screen and calls the API directly can ask for the whole table. Two typical states cause it:
- RLS is switched off on a table.
- RLS is on, but a policy is broader than intended, for example one that allows any request, or one that does not match rows to the signed-in user.
The test to run on your own app today
This is the same test as step 2 of our guide to rescuing a broken Lovable app, repeated here because it is the one that matters this week. Use two accounts that belong to two made-up users, never real customers.
- Create some data as user A.
- Sign in as user B and try to read, edit and delete A's data through the app.
- Try the same with no sign-in at all.
- Repeat for every table that holds user data, including file storage buckets.
Passing that through the app is not enough, because the screen may hide the problem. Test past the app. Sign in as B, copy B's access token from the browser's developer tools (local storage, the entry named like sb-...-auth-token), and ask the API for the whole table:
curl "https://<project>.supabase.co/rest/v1/<table>?select=*" \
-H "apikey: <anon key>" \
-H "Authorization: Bearer <B's access token>"
Any row that is not B's means the policy, or the missing RLS, is the problem, whatever the app shows. Run it once more with the anon key in both headers to cover the no-sign-in case. Do this for every table, not only the one you think is sensitive.
A blocked request is often silent. With RLS on and no matching policy, a read comes back as an empty list and an update changes 0 rows, and neither raises an error; only an insert fails, with error 42501. So "users see nothing" can be a missing select policy. List the table's policies in the SQL editor:
select * from pg_policies where tablename = '<table>';
Also check which tables have RLS at all, and look at your storage buckets, because a public bucket serves its files to anyone with the link. This is a check, not a certification. Passing it does not prove the app is secure.
If rows leaked
If the test returned rows that should not be visible, work in this order:
- Turn on RLS for the table, and write policies that match rows to the signed-in user. Test again with the same requests, because a badly written policy can lock everyone out, which looks like a separate bug.
- Retest past the app for every table and bucket, with B's token and with the anon key only.
- Rotate exposed keys. If a service_role key or any other secret was ever in the browser code, replace it at the provider and delete the old one. Removing it from the code is not enough, because it can stay in git history (why).
- Check your logs and exports for signs of bulk reads, as far as your plan keeps them.
- Think about the people in the data. If personal data of other people was exposed, check your obligations. For apps with EU users, the GDPR has notification duties with a short deadline (72 hours for reporting to the authority in many cases), and a data protection professional or lawyer should tell you what applies to you. This article is only a pointer.
Before any change, take a backup of the database and a copy of the code, so a policy fix cannot cost you data.
When to get help
If you can run the test and fix the policies yourself, do. If you are not sure the result is right, or the app holds sensitive data, a second pair of eyes is cheap compared with an incident. Devaland offers AI App Rescue, a fixed-price diagnosis for apps built with Lovable, Bolt, Cursor and similar tools. As read on the page on 15 November 2026: the diagnosis is 390 EUR excluding VAT, with a written report, and a fixed price for each fix, and the code stays yours. You export the code to GitHub with read access; we never log into your Lovable account. The report arrives within 5 working days after we receive access, and each fix is a separate fixed price agreed in writing before work starts. The page lists fintech and payment-processing apps and apps that hold health data as things we do not take, and it says we do not promise that every bug is found or any result. Check the page for the current terms before you order.
Whatever you choose, the researchers' own advice applies: review access permissions before an app handles real customer data, and repeat the review after significant changes.
FAQ
Is my Lovable app leaking data?
I cannot tell from here, and neither can a news report. The research is a sample of public apps and says so itself. The only way to know about your app is to test it: ask the database API for the whole table as another user and with the anon key only.
Does Lovable's scan replace the test above?
Lovable's head of security said a pre-publish scan and a free deeper scan exist. The researchers at Crissy Field argue that automated checks are no substitute for a professional security assessment. I would use any scan as one input and still run the direct API test yourself, because it is quick and answers the exact question.
Is the Supabase anon key a secret?
No. It is designed to be in the browser, which is exactly why every table it can reach needs RLS. A service_role key is different: it bypasses RLS and must never be in browser code.
Why does my app look fine if the table is open?
Because the screen filters rows by the signed-in user. The filter is a request the browser makes, not a rule the database enforces. Someone calling the API directly does not have to include the filter.
Do I have to tell my users if rows were exposed?
That depends on what was exposed, where your users live and whether the data was actually accessed. Check your obligations with a data protection professional or lawyer, and do not rely on this article for it.
Can Devaland fix my app?
We can diagnose it and quote each fix at a fixed price, within the limits on the AI App Rescue page. We do not promise results, and we do not take fintech or health-data apps.
Originally published on devaland.com: Lovable data exposure (November 2026): what was found, and the RLS test to run on your own app today
Top comments (0)