Is Bolt.new safe?
Is Bolt.new safe to ship?
Bolt.new builds and runs a full-stack app entirely in the browser using StackBlitz WebContainers, then deploys it in a couple of clicks. That speed is real. So is the gap it tends to leave in the database it wires up.
What Bolt actually does well
Runs a real dev environment, frontend and backend, directly in the browser with no local setup.
Connects cleanly to Supabase, Netlify, and Vercel for a working deploy in minutes, not days.
Handles multi-file edits across the stack without losing track of what already works.
The gap
The gap: RLS-off tables and env vars marked public without realizing it
When Bolt wires up Supabase, the same default applies as everywhere else in this space: tables ship without row-level security unless someone explicitly turns it on. On top of that, Bolt's in-browser environment variable settings make it easy to mark a variable as exposed to the client without registering that "exposed" means it ships in the JavaScript bundle anyone can read. A secret key set that way is no different from pasting it directly into your HTML.
Check Bolt for this yourself right now
Two checks, about 60 seconds combined.
- 01
Open your live Bolt app, open devtools, and go to the Network tab. Find a request to a supabase.co domain and copy the apikey header value.
- 02
Run: curl "https://<your-project>.supabase.co/rest/v1/<table-name>?select=*&limit=1" -H "apikey: <anon-key>" and see if rows come back with no login.
- 03
Separately, open devtools Sources tab, search the main JS bundle for "sk_live", "sk-", "AKIA", or "service_role".
- 04
If any of those patterns show up in a file that ships to the browser, that key is public and should be rotated immediately.
Rather not do this by hand? The free Leak Check runs the anon-key probe and a client bundle scan against your live URL and gives you results in under a minute.
Run the free Leak Check →Where the numbers come from
We scanned 66 live apps built with Lovable, Bolt, Cursor, base44, and Tempo. 41% of the Supabase-backed apps had at least one table anyone could read with the public anon key. The full methodology and findings are in the data report.
Read the data report →