Smart enough to trust, bounded enough to be safe
AnVerified's models exist to spend friction where risk is real and remove it everywhere else — while the protocol's hard refusals keep the intelligence from becoming the surveillance.
The math behind the yes/no
On-device liveness, on-device match
Biometric comparison runs where the face is — the phone. The vault receives a signed match result, never an image. There is no central face database to breach because the architecture refuses to build one.
Risk-adaptive re-checks
Silent annual re-attestation by default; step-up only when signals genuinely change (new device + new scope + unusual hour). Verification that nags constantly is verification people bypass — friction is spent where risk actually lives.
Range proofs, not records
A lender needs 'income band B', a benefit scheme needs 'below threshold'. Zero-knowledge-style range claims let the source prove the band without releasing the number — the PDPA-minimum made computable.
Credential-abuse detection
Issuance farms, registrar-account takeover and screenshot-forwards die here: issuance velocity models, attestation-chain checks and short-lived rotating tokens make a leaked credential expire before it travels.
Deepfake-resistant step-up
Video-KYC step-ups use challenge protocols tuned against generation models — the honest posture: assume synthetic identity attempts, make them expensive, keep humans in the loop for the edge.
Attestation routing intelligence
Which source can prove which claim, at what assurance, for what scope — the directory itself is learned and scored, so a relying party always gets the strongest available evidence path, never the cheapest.
What the models will never do
No emotion, no traits
Models decide validity and risk — never 'confidence in your character'. Identity intelligence that grades humans is not shipping here.
Explainable denial, always
Every failed verification states its reason and its appeal path in plain language. An unexplainable no is a bug with a victim.
The person holds the stronger model
Revocation, scope inspection and 'who asked' history belong to the human, readable at a glance. Transparency is the counterweight to the machinery.
Ready to build something intelligent?
Let's engineer your next product — from idea to launch.