Safety
Safe by design
We only actively test sites you prove you own. Without proof, ScanAnt stays passive and reads only what your site already serves to everyone.
Ownership first
Active checks are blocked on the server until ownership is verified, not just hidden in the interface.
Nothing destructive
No deleting or changing your data, no password guessing, no load tests. Anything that could leave a trace is stated before you start.
Secrets stay masked
Anything we find is stored only as a mask and a location, never in full, and never sent to an AI model. Data values are not saved.
Polite and logged
A clear User-Agent, a request budget per scan, and an audit log of every scan.
Rules
The rules the scanner follows
These rules are built into the scan engine on the server, not just written on this page.
Public addresses only
Only http(s). Private, internal and cloud-metadata addresses are blocked, including after redirects and DNS tricks.
Ownership checked right before
Before every full scan the server checks your verification again, so a removed DNS record stops deep checks at once.
Secrets masked, never used
Found keys are stored only as a mask and a location. They never go to logs or to an AI model, and we never call anything with them.
Data: proof, not copies
At most one or two rows are read as proof. We keep the fact of access, table and column names and row counts — not values.
Nothing destructive
No changing or deleting data, no password guessing, no exploits, no form submissions, no load tests.
Limits and an audit log
Our own User-Agent, a request budget and limited parallelism per scan. Every scan is written to an audit log.
Extra probes are opt-in
Deeper probes run only for verified owners who tick them, and anything that may leave a trace is named before the run.
Opt out any time
Site owners can ask us to block their domain from being scanned through the opt-out form.
Why it matters
Why it matters
AI builders ship fast, and security is often the thing that gets skipped. The figures below come from scanner vendors' own research, not from independent audits, so treat them as indicative.
- 39%
of 359 scanned Supabase apps (141) had exposed tables or row-level security errors.
Vendor-reported: Vibe App Scanner
- 12 in 100
scanned apps sent a Supabase service_role key to the browser.
Vendor-reported: vibecheck
- CVSS 9.3
Severity out of 10
CVE-2025-48757: ineffective row-level security put 170+ Lovable-built apps at risk.
Public CVE record
Vendor-reported numbers come from the vendors' own samples and methods and have not been independently verified.