LynxPay
Pre-payment social-engineering defence

The UPI PIN proves it's you. It doesn't prove it's your idea.

LynxPay is a pre-payment security layer that catches socially engineered UPI scams before authorization, by checking whether a payment still matches the user's genuine intent.

Synthetic demo · no real UPI, bank, telecom or device
LynxPay AI guardian
Message received · hi-en
Bijli ka bill update nahi hua. Aaj raat light kat jayegi. Ye QR scan karke turant ₹4,999 bharo.
Scan & pay ₹4,999
QR from message
0
false positives
on our adversarial-benign benchmark v1
0.99
F1 score
scam detection, benchmark v1
8
languages
incl. Hindi, Hinglish, Tanglish, Bengali, Marathi, Telugu
14
live scenarios
run any of them in two clicks
Agentic AI, governed

An AI guardian on every payment. It advises; a transparent policy decides.

Step 1

Perceive

The AI guardian gathers the signals around a payment: the message or call, the QR source, the beneficiary, the amount and the user's history.

Step 2

Understand

It reads what the user thinks they're paying for, in their own language, and the recipient that purpose implies.

Step 3

Reason

It checks the perceived purpose against where the money is actually going, and scores behaviour, network and rules.

Step 4

Act before the PIN

Allow, warn, step-up or block, with a plain-language reason. Every verdict is signed, logged and open to human override.

Only an intent layer can do this

What our AI can do that a scam filter can't

It catches AI-powered scams too. A deepfake "digital arrest" voice and a prompt-injected payment agent are the same failure: a payment that is authorized but not intended. The guardian blocks both, live, in the demo.
Responsible AI

Built to be trusted, and to be audited

Explainable verdicts

Every decision ships with human-readable evidence and a full audit trail. No black box.

The AI never decides

The model only understands language. A transparent policy makes the call, so a regulator can sign off on it.

Human oversight

An analyst can review and override any decision, and the reason is recorded.

Privacy & sovereignty

Synthetic today; the production design is consent-based and can run on a self-hosted, in-region model.

Multilingual: English, Hindi, Tamil, Hinglish, Tanglish, Bengali, Marathi and Telugu.
Where it fits

It sits on top of the rails, and sees what they can't

Device & account
UPI PIN
Bank / PSP fraud rules
LynxPay intent layer
Authorization

Existing controls verify the person and the pattern. LynxPay verifies the purpose.

Built to integrate

A pre-authorization API, not another app to install

LynxPay drops into the payment flow a bank, PSP or UPI app already runs. One call before the PIN returns a decision, a plain-language reason, and a signed record.

POST /v1/risk/evaluate

Pre-authorization check

A wallet, PSP or bank calls LynxPay before confirming a payment. Back comes Allow / Warn / Step-up / Block, the evidence, and a signed intent record, in milliseconds.

POST /v1/agent/assess

Agent intent-firewall

AI-agent and UPI-Circle-style platforms check a payment against its mandate before it executes, so a prompt-injected agent can't overpay.

POST /v1/intent/verify

Public verification

A bank's dispute desk or an ombudsman verifies a signed intent record against the public key, with no access to our systems.

Deploy where the data lives. The API is stateless, provider-swappable and self-hostable, designed to run in-region inside a bank or PSP so payment data never leaves the deployment. Synthetic today; the same contract is built for a consented pilot.

Watch a scam get caught before the PIN.

No sign-up. Preloaded scenarios. Two clicks to a live verdict.