VerityGraph
Fraud detection and loss control for Indian banks.
The problem
Fraud is moving faster than the institutions defending against it. AI has made impersonation cheap: deepfaked video calls, cloned voices, fabricated KYC. UPI and instant rails settle value in seconds, with no batch window to review it.
The fraud running on instant rails has no batch window to learn from. Rules written for yesterday’s fraud will not recognise a mule network fanning money across forty accounts in four minutes.
Indian banks carry the same obligation as the largest global institutions with far fewer staff — and under the direct gaze of the RBI and FIU-IND.
The product
One graph of the bank’s own data: accounts, customers, devices, sessions, counterparties and outbound destinations, across fraud, cyber and AML. A pattern that is invisible in one table becomes a shape across many.
What lands on an investigator’s desk is one case with one timeline, graded Confirmed, Probable or Lead-only, tagged to named MITRE F3 techniques and structured for FIU-IND reporting. What lands on the board’s desk is a periodic report of actual fraud loss against a limit the bank set in advance.
Every target is measured against your own baseline, agreed in writing before work begins.
Sources: Ministry of Home Affairs / I4C reply in Lok Sabha, July 2025; NCRB Crime in India 2024, May 2026.
How it runs
Your own instance
Analytics run inside hardware-enforced enclaves, in your own data centre or cloud tenancy, and you verify the attestation yourself. Customer data, transaction data and case files never leave India.
A shared clean room
Only adversary indicators reach a shared clean room, so what one bank learns about an attacker’s infrastructure protects the next without exposing anyone’s customers.
Full capability alone
Detection and forensics work on day one with a member count of one. The clean room adds value as it grows; it is never a prerequisite.
Four capabilities, one graph
Measure the real cost
Most banks record fraud as a charge-off. VerityGraph records what each case actually cost, including the losses a charge-off never shows, and after a year gives you your own baseline and cost multiplier instead of an industry average.
Set a limit and hold to it
Credit risk has an appetite and a committee. Fraud usually has a report. Once the baseline exists, the bank states what it will lose this year; VerityGraph tracks it and tightens controls automatically on terms agreed in writing beforehand.
Carry the team’s workload
Agents triage alerts, assemble case files, draft narratives and prepare regulatory output. Every case ends in human review; nothing ends in an automated adverse action.
Share on authority you already hold
Structured intelligence for FIU-IND reporting and law-enforcement referral, with referral packages built around a criminal scheme rather than a single victim, so a case clears a prosecutor’s threshold.
How banks engage
Proof of concept, 8 weeks, at cost
Your data only, one typology, no sharing agreement and no production integration. Success thresholds agreed in writing against your own measured baseline before work begins.
Production pilot, 12 to 16 weeks, fixed fee
Live traffic inside your perimeter, operated by your own team, with a loss budget agreed at the outset and reported against for the duration.
Implementation, at your pace
The pilot fee is credited in full against the implementation contract. Network participation is optional and never a prerequisite.
You can hold an employee accountable. You cannot fire an AI agent.
Agentic AI runs at machine speed and does not rest, on both sides of the wall. VerityGraph is built for the side that has to keep up.
Tell us what you are facing
Banks, energy operators and research programs are all welcome.