In short
- Start with the database rules. In the scan behind CVE-2025-48757, 170 of 1,645 apps built with Lovable had tables that strangers could read or write.
- A public key in your bundle is normal. A secret key there is a leak, and removing it from the code does not undo it: the key has to be revoked.
- Test with two accounts. Most data leaks between users are invisible when you only ever sign in as yourself.
- Before a launch, set spending caps at every provider that bills per call. A cap turns a bad night into a fixed loss.
- Compliance badges, a bug bounty and a perfect header score can wait until someone asks for them.
On this page
- What actually gets a solo product in trouble
- 1. Open data rules: check these first
- 2. A secret key in the bundle
- 3. One user reading another's data
- 4. An AI route anyone can call
- 5. A webhook that accepts unsigned events
- Before you launch on Product Hunt or Hacker News
- What can wait
- What Keybrum does for a solo maker
- Questions
- Sources
What actually gets a solo product in trouble
Security advice for startups is mostly written for teams: policies, audits, access reviews. If you build alone, most of it does not apply, and the part that does is short. You have one app, one database, a few API keys and no time.
The public cases are consistent about where small products leak. In 2025, a researcher tested 1,645 apps built with Lovable and found 170 with database tables that anyone could read or write, about one in ten. That became CVE-2025-48757. In January 2026, the social network Moltbook went viral, and within days researchers at Wiz found that its whole database answered the public key in its JavaScript: 1.5 million API tokens and about 35,000 email addresses, as told in the Moltbook data leak guide. Its founder had said he did not write a line of the code.
Neither case involved breaking anything. The app worked, the demo was fine, and a setting that decides who can read the data had never been written. That is the pattern to look for, in five places.
| Order | The gap | What a stranger gets | Time to check |
|---|---|---|---|
| 1 | Open data rules | Your whole database or your users' files | 10 minutes |
| 2 | A secret key in the bundle | Your account at that provider, and its bill | 10 minutes |
| 3 | One user reading another's data | Every customer's records, one ID at a time | 20 minutes |
| 4 | An AI route anyone can call | Your model budget | 5 minutes |
| 5 | An unsigned webhook | A paid plan without paying | 5 minutes |
1. Open data rules: check these first
If your frontend talks to Supabase or Firebase directly, the key in your JavaScript is public by design. What protects the data is a rule on each table or collection: row level security on Supabase, security rules on Firebase. A table without one answers anyone.
-- Supabase: tables in the public schema with row level security off
select tablename
from pg_tables
where schemaname = 'public' and not rowsecurity;Then ask your own API as a stranger: request one row of a private table with only the public key. An empty list or a permission error is what you want. The Supabase RLS guide has the policies to paste. On Firebase, look for if true and for rules that stop at request.auth != null, and see the Firebase security rules guide. Do the same for storage buckets: the Tea app breach exposed about 13,000 verification selfies and ID photos held in old storage.
2. A secret key in the bundle
Some keys are meant for the browser: the Supabase publishable key, the Firebase config, the Stripe publishable key. Others give full access to an account: the Supabase service role key, an OpenAI or Anthropic key, a Stripe secret key. The second kind ends up in the bundle when it is stored in a variable whose name starts with NEXT_PUBLIC_ or VITE_, because the build copies those into the JavaScript every visitor downloads.
# After a production build, from your project folder.
# Adjust the folders to your framework: dist, build, .next/static, out.
grep -rEo "sk-[A-Za-z0-9_-]{8}|(sk|rk)_(live|test)_[A-Za-z0-9]{8}|sb_secret_[A-Za-z0-9]{8}|whsec_[A-Za-z0-9]{8}" dist .next/static 2>/dev/null | sort -ueyJ: paste it into the Supabase key checker to see which one it is. No output is what you want.If you find one, the order matters: create a new key kept on the server, deploy, revoke the old one, then check the provider's usage page. Deleting the key from the code changes nothing for the copies already downloaded. The leaked API key guide gives the steps per provider, and the API key leak checker names every key in a pasted bundle or .env file.
3. One user reading another's data
This is the gap you cannot see alone. Signed in as yourself, every page shows your own data and everything looks right. The question is what happens when user B asks for user A's invoice by changing an ID in the address or in a request. The name for it is IDOR, for insecure direct object reference.
- 1
Create two test accounts
Use two browsers, or one normal window and one private window. Create a record in account A: a project, a note, an invoice.
- 2
Copy A's identifier
Take the ID from the address bar or from the network tab of the developer tools.
- 3
Ask for it as B
Signed in as B, open the same address, or resend the same API request with B's session. Try reading, then editing.
- 4
Read the answer
A 403 or 404 is correct. A's data on B's screen means every user can read every other user's records.
The fix is an ownership check on the server for every route that takes an ID, or a database rule tied to the signed-in user. The IDOR guide shows both. Check the fields a user can set on their own profile too: if role, plan or credits is writable from the client, upgrading is one request.
4. An AI route anyone can call
If your product calls a model, there is a route such as /api/chat or an Edge Function that forwards prompts to OpenAI or Anthropic with your key. If that route does not check for a session, anyone can use it as a free API, and the bill is yours.
# Your own app only. No cookie, no token.
curl -s -o /dev/null -w "%{http_code}\n" \
-X POST "https://YOUR-APP.com/api/chat" \
-H "content-type: application/json" \
-d '{"messages":[{"role":"user","content":"hello"}]}'Require a session first, then add a per-user limit and a daily cap, and set a hard spending limit in the model provider's dashboard. The unprotected AI endpoint guide goes layer by layer, and the rate limiting guide covers the limits on common stacks.
5. A webhook that accepts unsigned events
When someone pays, your payment provider calls a webhook route on your server, and your code upgrades the account. The provider signs each event. If your route does not verify the signature, anyone who finds the URL can send a fake "payment completed" event and get the paid plan.
The check: send an empty POST with no signature header to your own webhook route. A 400 that mentions the signature is correct. A 200 means the route trusts whatever arrives. The Stripe webhook security guide has the verification code and the test, and the Stripe security guide covers the keys and the price.
Before you launch on Product Hunt or Hacker News
A launch changes who looks at your app. For months the only visitor was you. On launch day it is thousands of people, many of them developers, and some will open the developer tools out of habit. Moltbook was examined within days of going viral. Run through this list the day before.
- The five checks above, run on the production app
- A spending cap or a billing alert at every provider that bills per call: models, email, SMS, maps
- Sign-up needs an email confirmation or a challenge before an account can use anything that costs you money
- Two-factor sign-in on your hosting, database, domain registrar, payment and GitHub accounts
- Your framework is on a patched version: the Next.js CVE checker takes a
package.json - A security.txt file with an address you read, so the person who finds something can tell you
- A recent database backup that you have restored at least once
- No
.envfile, source map or.gitfolder served by the site: see exposed .env files
If the launch goes well, you will ship fixes all day. Each deploy can reopen one of the five: a new table without a policy, a new route without a check. Run the checks again after the day's last deploy.
What can wait
Time is the constraint, so it helps to know what to skip for now. None of the items below protects your users' data on launch day.
| Can wait | Until |
|---|---|
| SOC 2, ISO 27001 and other certifications | A customer's procurement team asks. See security questionnaire answers for the first requests. |
| A manual penetration test by a firm | You sell to companies that require one, or you handle regulated data. The startup pentest guide helps decide. |
| A bug bounty programme | You have the time to answer reports within days. A contact address is enough before that. |
| A top grade on security headers | After the five gaps. Headers are a second layer, covered in the security headers guide. |
| A web application firewall tuned by hand | You see abuse that a simple rate limit does not stop. |
| Written security policies | You hire, or a customer asks to read them. |
Two things do not belong in this table even though they feel optional: backups and two-factor sign-in on your own accounts. They take minutes, and nothing else helps once you need them.
What Keybrum does for a solo maker
Keybrum is the automated pentest we build for apps made with AI. It runs the checks on this page, and the ones that take longer than ten minutes, against your live app:
- it asks every Supabase table, RPC and bucket named in your JavaScript for data with only the public key, then as a new account, and does the same for Firestore, the Realtime Database and Firebase Storage;
- it looks for secret keys in the bundle and in API answers, and checks with a free read-only call whether each one still works;
- it creates two accounts, has the first create a marked record, and tries to read and change it from the second;
- it sends three short prompts to routes that look like AI endpoints, without a session;
- it sends one forged, unsigned event to webhook routes, with identifiers that match nothing;
- it checks for exposed
.envfiles, source maps and.gitfolders, and for known Next.js and React flaws by their behaviour.
Every finding comes with the request that worked, personal data redacted, and a fix written for your coding agent. After the fix, the same attack runs again, and the finding closes only when it fails.
What it does not cover: your own accounts and laptop, your backups, phishing against you, and anything that needs denial-of-service traffic. It never tests deleting data. It only attacks apps whose owner proved ownership, so it cannot be pointed at someone else's product.
The check at /check is free and needs no account. A one-time report costs €49: one app attacked once, with retests of its findings for 7 days and a PDF. The Maker plan is €29 a month excluding VAT for 5 apps, and attacks each of them again on every deploy, which is when the five gaps come back.
Questions
What security does an indie hacker really need?
Five checks cover the gaps behind the public cases cited on this page: database and storage rules, no secret key in the bundle, no user able to read another's data, no AI route open to strangers, and signed webhooks. Add spending caps, backups and two-factor sign-in on your own accounts.
How long does it take to check a micro SaaS before launch?
About an hour for the five checks on this page if you know your stack, most of it spent on the two-account test. Fixes range from one SQL statement to an afternoon of adding ownership checks to routes.
Do I need a penetration test as a solo founder?
A manual test by a firm is rarely needed before you sell to companies that ask for one. What you need earlier is to know what a stranger can do to the live app, which you can check by hand with this page or with an automated test.
Is it safe to launch an app built with Lovable, Bolt or Cursor?
It can be, and the tool is not the deciding factor. Generated apps work on day one with the database open, because an open table passes every demo. Check the rules, the keys and the routes on the deployed app before the launch.
What is the most common security mistake in indie SaaS?
From the public cases, database tables without access rules. The scan behind CVE-2025-48757 found it in 170 of 1,645 Lovable apps tested, and the Moltbook leak had the same cause.
What should I do if I find a leak after launch?
Close the access first, rotate any exposed key, then find out what could be read and since when. The guide on what to do when your app is hacked gives the order for the first hour, day and week, including who to tell.
How much does Keybrum cost for a solo maker?
The check is free. A one-time report on one app costs €49. The Maker plan is €29 a month excluding VAT for 5 apps, with every deploy attacked again.
Sources
- 1.Statement on CVE-2025-48757, Matt Palmer
- 2.Hacking Moltbook: AI Social Network Reveals 1.5M API Keys, Wiz Research
- 3.Row Level Security, Supabase Docs
- 4.Fix insecure rules, Firebase Docs
- 5.API keys, Stripe Docs
- 6.Hackers leak 13,000 user photos and IDs from the Tea app, designed as a women's safe space, NBC News