Introduction
India’s real-time digital payments architecture, spearheaded by the Unified Payments Interface, has achieved unprecedented operational scale, handling over 750 million daily transactions with a strategic target of exceeding one billion daily transactions. However, this explosive growth in transaction volume has coincided with an increasingly sophisticated landscape of financial cybercrime, forcing regulatory authorities and infrastructure operators to fundamentally re-evaluate fraud detection mechanisms. As banks face heightened regulatory scrutiny and stricter compliance rules, the integration of Artificial Intelligence has transitioned from an optional operational upgrade to a core legal and supervisory imperative.
To navigate this transition, the Reserve Bank of India and the National Payments Corporation of India are deploying a dual approach that combines advanced algorithmic surveillance with stringent liability frameworks. This regulatory push seeks to balance the rapid expansion of digital financial inclusion with absolute operational resilience. For banks, payment aggregators, and fintech entities, the emerging mandate requires moving away from generic technological solutions toward specialized, rule-bound systems capable of defending real-time payment networks against systemically damaging fraud.
Small Language Models and Deterministic Architectures in Payments
A central pillar of NPCI’s technological strategy is the deliberate shift toward smaller, process-tied language models rather than massive, open-ended general-purpose systems. National Payments Corporation of India Chief Executive Officer Dilip Asbe emphasized that the payment industry’s primary opportunity lies in constructing models that are sharp, highly specific, and strictly deterministic.
In financial operations, especially real-time consumer support and dispute resolution, open-ended artificial intelligence models introduce severe operational risk if they provide ambiguous, non-standard, or hallucinatory answers.
To illustrate this approach, NPCI has operationalized FiMI, a domain-specific finance model designed explicitly for payment troubleshooting and customer support. Serving over one million users for tasks such as automated mandate cancellations and real-time transaction issue resolution, FiMI demonstrates the efficacy of narrow artificial intelligence embedded directly within operational workflows.
By tying algorithms to rigid, auditable process rules rather than open-ended dialogue, payment networks can scale user support and maintain deterministic reliability without exposing financial institutions to procedural or legal liability. scale user support and maintain deterministic reliability without exposing financial institutions to procedural or legal liability.
Tightening Regulatory Timelines and Statutory Fraud Compensation
Parallel to technological upgrades, the Reserve Bank of India is significantly raising the regulatory cost of inadequate fraud prevention through a comprehensive framework taking effect on January 1, 2027. Under these upcoming rules reported by the Economic Times, regulated financial entities face strict statutory deadlines for investigating and resolving digital fraud complaints. Banks will be legally obligated to investigate and resolve domestic fraud claims within 45 calendar days, while cross-border fraud disputes must be concluded within 60 calendar days. In addition to accelerated dispute resolution timelines, the regulatory framework establishes a
mandatory victim compensation mechanism for smaller digital fraud losses. Customers suffering fraud losses of up to fifty thousand rupees, equivalent to approximately five hundred and twenty-nine dollars, may be eligible to receive a once-in-a-lifetime compensation of up to twenty-five thousand rupees directly from banks. By establishing direct financial liability for unaddressed fraud, the Reserve Bank of India has created a compelling economic incentive for banking institutions to deploy real-time, highly accurate fraud identification tools.
Combatting Cyber Fraud Through MuleHunter.AI and Real-Time Surveillance
Addressing the systemic threat of illicit money movement, the Union Finance Ministry has instructed banking institutions to adopt MuleHunter.AI, a specialized artificial intelligence tool developed by the Reserve Bank of India specifically to detect and neutralize mule accounts. Mule accounts, which are frequently used by cybercriminals to layer and launder funds obtained through digital fraud, represent a crucial vulnerability in real-time settlement networks. Traditional, retrospective auditing mechanisms are fundamentally incapable of stopping high-velocity transfers through these accounts.
By deploying real-time surveillance tools like MuleHunter.AI, financial institutions can identify anomalous account behavior, flag suspicious opening credentials, and freeze illegal fund flows before money is siphoned out of the banking network. This mandate reflects a broader industry recognition that effective fraud screening requires narrow, process-driven algorithms integrated directly into onboarding and transaction monitoring pipelines, ensuring full auditability under regulatory inspection.
Significance for Banks, Payment Operators, and Fintechs
The convergence of strict regulatory liability and real-time algorithmic surveillance fundamentally alters the operating environment for Indian financial institutions. Payment system operators can no longer rely solely on post-facto dispute mechanisms or manual fraud reviews. The impending 2027 Reserve Bank of India framework shifts the burden of proof and financial liability onto regulated entities, making real-time, pre-transaction screening an operational necessity.
Furthermore, as India attracts significant global artificial intelligence infrastructure investment, including OpenAI’s reported one hundred megawatt presence, the domestic financial stack is becoming a global test case for regulated operational AI. Entities that successfully integrate auditable, deterministic artificial intelligence into dispute management, user onboarding, and mule detection will not only minimize regulatory exposure but also set the benchmark for real-time payment governance internationally.
AMLEGALS Remarks
The Reserve Bank of India’s 2027 framework and NPCI’s deployment of deterministic artificial intelligence mark a fundamental shift from voluntary fraud mitigation to strict legal accountability. By establishing rigid forty-five-day and sixty-day dispute resolution windows alongside mandatory compensation thresholds for fraud victims, the central bank has placed the financial and legal burden of cyber fraud squarely on payment fiduciaries. In this heightened regulatory climate, generic or non-deterministic artificial intelligence models present severe compliance risks, as non-standard or delayed dispute workflows will directly trigger statutory penalties, compulsory compensation payouts, and regulatory sanctions. Financial institutions must recognize that fraud management is no longer merely an operational loss-mitigation function, but a core statutory compliance requirement.
Furthermore, the mandatory implementation of specialized tools such as MuleHunter.AI and NPCI’s small language models demonstrates that regulatory compliance in high-volume payment networks requires specialized, process-tied technology. Banks and fintechs must pivot away from broad, general-purpose generative tools toward narrow, fully auditable algorithms that operate within clear, deterministic rules. As India’s payment stack approaches one billion daily transactions, financial institutions that fail to integrate auditable artificial intelligence into their fraud screening and dispute resolution mechanisms will face mounting regulatory exposure, operational bottlenecks, and substantial financial liabilities under incoming Reserve Bank of India directives.
For any queries or feedback, feel free to connect with Hiteashi.desai@amlegals.com or Khilansha.mukhija@amlegals.com