Kinyarwanda
Oshiwambo
Tagalog
Punjabi
Proto has been selected as a finalist in the AI-driven Fraud & Scams problem space of the Cambridge Digital Innovation and Regulation (C:>DIR) hackathon, with a proposal developed alongside the National Bank of Rwanda and fund-tracing partner FNA.
The submission proposes a set of AI agent network connectors for a proposed anti-scam initiative for Rwanda’s National Digital Payment System (RNDPS/eKash). Given the rise in AI-based fraud, the proposal uses the Agentic Regulator sandbox to build a modern AI agent architecture for grievance intake, redress and mediation, fund tracing, recovery, supervisory insights and fraud detection.
Why Rwanda, why now
TransUnion’s H1 2025 Rwanda fraud report found that 67% of respondents had been targeted by email, online, phone-call or text-message fraud in the prior three months, with 10% saying they had become victims. Money-mule scams and social-engineering scams were the two most frequently reported methods, at 30% and 29% respectively. Rwanda’s RNDPS/eKash is already operating at national scale – by June 2025 it had 2.16 million active users and recorded 25 million transactions worth FRW 105 billion in that month alone – underlining the importance of fast, coordinated scam reporting and fund tracing as usage grows.
The proposal builds directly on Intumwa, the National Bank of Rwanda’s AI agent for consumer protection, developed by Proto and already handling complaints and information in Kinyarwanda, English and French across webchat and WhatsApp for more than 600 financial institutions – giving the initiative a proven front door for scam reporting rather than a system built from scratch.
The fund-tracing and recovery side of the proposal draws on FNA’s fraud-network mapping technology, the same capability behind Proto and FNA’s G20 TechSprint award for cybercrime and fraud mitigation, won in November 2025 for their joint submission Trust at Speed: A National Utility for Scam Reporting and Fund Tracing. In Malaysia, FNA’s National Fraud Portal lifted fund-freezing rates from 0.5% to 21.5% for promptly reported cases, with full recovery achieved when victims reported immediately; Bank Negara Malaysia separately reports a 75% reduction in the time needed to trace stolen funds – the benchmark the Rwanda proposal aims to replicate.
How the agents work together
In the proposed design, a grievance intake agent – handling Kinyarwanda, English and French, by text and voice – triages and lodges a scam report in under a minute via Intumwa. A fund-tracing agent then maps the flow of funds across accounts and coordinates a freeze request, escalating to a cross-border bridge if funds have moved abroad. A recovery agent drafts a referral and evidence packet for Rwanda’s Financial Intelligence Centre to decide on freeze or seizure, while a supervisory insights agent aggregates anonymised cases into typologies for the National Bank of Rwanda’s fraud-prevention work – with a further workflow dedicated to detecting emerging fraud patterns. Every agent’s output is checked by a verifier agent before reaching a human.
Autonomy is bounded: agents can read, triage, trace, draft and alert on their own, but every irreversible step – freezing an account, authorising reimbursement, requesting a cross-border freeze, issuing a public warning – requires named human sign-off. Tool access is permissioned throughout, so the payment-system connector, for example, can flag a transaction but never move or debit funds.
This system will be prototyped in the Agentic Regulator sandbox, and if successful, the National Bank of Rwanda’s role as problem owner provides a path to operationalise elements of the technology within a future anti-scam initiative.
This capability also offers a regional blueprint for other countries facing similar consumer threats – addressing not only today’s scam challenges, but the growing threat of rapid, AI-based scams that move victim funds in minutes.