Payment systems are the circulatory infrastructure of the modern economy. They move value between households, firms, governments and financial institutions; underpin monetary policy transmission and contribute to financial stability. Central banks in particular have long recognised the safe and efficient operation of this infrastructure as a matter of fundamental public interest.
Yet, the frameworks through which authorities exercise oversight were built for a different era. The canonical model – periodic reporting, assessment against standards, dialogue with operators – was designed for systems of manageable complexity and moderate speed.
Today, new payment modalities and settlement instruments, such as fast payments, stablecoins, tokenised deposits and digital-currency instruments, are emerging alongside conventional infrastructures, while threats from fraud, cyber intrusion and operational contagion evolve faster than any static rulebook can accommodate. Decentralised platforms add further layers of complexity, operating without a single responsible operator and challenging the very notion of one or more oversight counterparts.
What has changed is not merely scale but structure. Modern payment ecosystems are faster, more interconnected, more dependent on complex technology chains and populated by a more diverse set of actors than legacy oversight frameworks were designed to govern. The sequential logic of traditional oversight is increasingly mismatched to this reality: complexity is outpacing capacity. Artificial intelligence can help close that gap, but only if it is embedded in a new oversight paradigm. We call it ‘techsight’.
A qualitative shift in oversight
Techsight should not be understood as the automation of existing oversight tasks. It is a qualitative shift in what oversight can perceive, anticipate and do. Traditional oversight operates on sampled, time-delayed information: an overseer receives periodic reports, conducts scheduled evaluations, analyses aggregated indicators and responds when deviations cross predefined thresholds. This model is inherently reactive as it sees the payment ecosystem through a rear-view mirror, at coarse resolution and with a lag that matters enormously when risks propagate across interconnected networks.
Techsight inverts this logic. Machine-learning models operate on granular transaction and operational data streams, learn the baseline distribution of system behaviour and flag deviations as they emerge. They model interdependencies between participants, trace potential contagion paths and identify concentrations that bilateral or aggregate reporting obscures. The objective is not simply faster reporting; it is continuous systemic intelligence.
For example, anomaly-detection models surface spikes in payment volumes, queue formation, settlement delays or liquidity mismatches. Natural-language-processing tools extract risk intelligence from incident reports and regulatory filings. Predictive analytics simulate contagion paths, allowing pre-emptive rather than post-facto intervention– oversight as continuous systemic intelligence rather than periodic audit.
The dual dynamic
The systems being overseen are themselves becoming deeply AI-reliant. Payment operators increasingly embed AI into fraud detection, transaction screening, liquidity management, customer onboarding and operational decision-making – a challenge policy discussions often underappreciate. Complex models produce decisions through layers of non-linear computation that are not readily intelligible to their designers, let alone to external authorities. This opacity weakens accountability. Where models are trained on biased and incomplete data, they may silently entrench discrimination or generate fragile operational behaviour.
Equally significant is systemic fragility from concentration. Where operators converge on a small number of vendors, data providers or model architectures, a vulnerability in one widely adopted solution can cascade across multiple institutions simultaneously. AI can therefore become both a source of operational efficiency and a new channel of correlated risk.
Yet, the same capabilities that operators deploy for efficiency can become instruments of oversight. An authority with appropriate access to granular data streams can form independent, near-real-time judgements, interrogate automated decisions, verify fairness, detect data drift and monitor model performance. This dual dynamic – AI-driven systems requiring AI-enabled oversight – defines the frontier of the techsight agenda.
Four non-negotiable governance principles
Capability requires governance. Four principles are non-negotiable.
First, explainability is not optional. When an oversight judgement is shaped by an AI model’s output, that output must be traceable, auditable and defensible. AI must be open to scrutiny; oversight cannot rest on sealed proprietary black boxes.
Second, human-in-the-loop is structural, not residual. AI tools should surface, prioritise and synthesise; highly specialised and regularly trained professionals must retain the authority to interpret, challenge, override and act. The risk of automation bias – where human reviewers defer uncritically to algorithmic outputs – must be actively managed through formal escalation paths and clear accountability.
Third, lifecycle governance is continuous. AI models drift and degrade as data distributions, fraud patterns, participant behaviour and operational environments change. A well-performing model may become misleading over time. Authorities must build capacity for validation, performance monitoring, retraining and independent challenge.
Fourth, institutional sovereignty matters. Oversight authorities must understand the tools they rely on, retain access to relevant data and preserve the ability to operate, interpret and challenge their analytical infrastructure independently. Outsourcing analytical judgement to opaque vendors is not a cost-saving measure; it is a transfer of regulatory authority. Technology partnerships are valuable, analytical dependency is not.
The institutional readiness gap
The greatest obstacle to AI-enabled oversight may not be technology itself but institutional readiness. Many authorities can procure advanced analytical tools. Far fewer possess the governance structures, data architecture, human capital, validation processes and accountability frameworks required to deploy them responsibly and effectively.
The transition to techsight demands more than technological investment. Authorities must assess whether their mandates, organisational structures, model-governance practices and staff capabilities are adequate for continuous, AI-assisted oversight – and how inspection planning, incident response and supervisory dialogue should change when risk signals are generated continuously rather than periodically.
Techsight is a governance and institutional-transformation agenda in which technology is the instrument. Its adoption should begin with a readiness assessment: what can the authority safely automate, what must remain judgement-based, where are the data gaps, which decisions require human escalation and how should accountability be allocated?
The techsight horizon
The risks associated with AI deployment in payments are transnational. International coordination is therefore structurally necessary. Common standards for techsight will become increasingly important, covering explainability, model validation, data governance, operational resilience and accountability.
Privacy-preserving data-sharing protocols may support collaborative risk intelligence without centralised data pooling. Capacity building will require sustained investment. The integrity of the global payment network depends on its weakest oversight link: risks originating in jurisdictions with weaker capabilities propagate through globally connected financial infrastructures.
The techsight agenda is about what kind of oversight the new world of payments requires, and what central banks and other relevant authorities must build, govern and continuously adapt to deliver it. The answer points towards an oversight function that is more analytically sophisticated, more continuously engaged and more internationally coordinated than today’s, capable of seeing the payment ecosystem as a whole, with a real-time understanding of its structure, dynamics and vulnerabilities.
That ambition is within reach, but it demands investment in technology, talent and governance, alongside institutional willingness to redesign operating models and collaborate across borders. Techsight is more than a collection of analytical tools. It is an emerging discipline for the governance of payment ecosystems, fit for the complexity of the world it must govern.
Biagio Bossone is an adviser to international financial institutions and national central banks. Claudio Ceresani is Chief Executive Officer and Co-founder of DGT Solutions and a global adviser on payment systems. Massimo Cirasino is CEO and Founder of the Payment System Academy and a global adviser on payment matters.
Interested in this topic? Subscribe to OMFIF’s newsletter for more.