Bunker AI
How Bunker AI works
Every verdict says what we checked, who checked it and what we didn't. If a check can't run, we say so instead of guessing.
- DangerousA trusted source lists it, or a strong signal fired: for example a Google Web Risk match or a signature match.
- SuspiciousNo trusted listing, but warning signs: a brand-new domain, a look-alike name, a request for gift cards or a one-time code.
- VerifiedAt least one trusted source checked it and found nothing, and every needed check ran. It is not a guarantee.
- Not verifiedOur basic checks found no warning signs, but no trusted source covers this kind of input yet.
- Couldn't checkA needed check is not set up, busy, or did not answer. We never turn that into Verified.
Sources
Google Web Risk
Lists of sites suspected of phishing, malware or unwanted software.
RDAP
Public domain registration records, used for the domain's age.
Anthropic
A language model that reviews message text for scam signals.
Sightengine
Scores images for signs of AI generation and face swaps.
Have I Been Pwned
Pwned Passwords, queried from your browser with only a hash prefix.
libphonenumber
Phone number validation and number types, run on our server.
Bunker signatures
The same file and agent signatures as our desktop app. The file list is small today.
Honest limits
- No checker catches everything. New scams appear every day.
- We do not open or render pages, so we can't see what a page shows.
- Our file signature list is small today: it proves the scanner works, it is not full antivirus coverage.
- AI-image scores are probabilities, not proof.
- Video deepfake checks, email breach lookups and carrier lookups are coming soon.