The institutions responsible for supervising the financial system are undergoing their own technology transformation. As banks deploy AI across fraud detection, credit assessment, compliance, cybersecurity and risk management, regulators are increasingly exploring the same technology to analyse institutions, identify unusual patterns and detect emerging risks. This creates a significant change in financial infrastructure: AI is no longer simply something regulators need to oversee. It is becoming part of the toolkit they use to conduct that oversight, raising the prospect of a banking system where increasingly automated institutions are monitored by supervisors equipped with increasingly intelligent analytical systems of their own.
The Bank for International Settlements has been drawing attention to this evolution. Fernando Restoy, Chair of the Financial Stability Institute, recently highlighted how financial supervisors are increasingly using AI to process supervisory information, identify emerging risks and make more effective use of supervisory resources. At the same time, banks are deploying the technology across areas ranging from fraud detection and credit assessment to compliance and risk management.
Those developments are often discussed separately, but their convergence could prove more significant. If banking becomes increasingly automated and operates closer to real time, supervisors may eventually need technology capable of observing and understanding risk at a similar speed.
Traditional banking supervision has relied heavily on information submitted at defined intervals. Banks collect data from multiple systems, reconcile it against regulatory definitions and produce reports that allow supervisors to assess areas such as capital, liquidity, credit quality and operational risk. That model remains essential, but it inevitably creates a gap between what is happening inside an institution and when a regulator receives a consolidated picture of it.
Supervisory technology, commonly known as SupTech, has the potential to narrow that gap.
The concept itself is not new, but the capabilities available to regulators are advancing quickly. AI can help analyse large quantities of structured and unstructured information, search regulatory documents, compare disclosures, identify inconsistencies and highlight patterns that warrant human attention. BIS research covering financial authorities across multiple jurisdictions has shown that supervisors are already experimenting with SupTech, although the transition from pilot projects to wider operational deployment remains uneven.
More sophisticated applications are also beginning to emerge. BIS researchers have explored the use of large language models to help supervisors compare financial documents with regulatory requirements and identify potential divergences for human review. This demonstrates how generative AI could eventually be used not simply to summarise information, but to help regulators interrogate large volumes of complex documentation that would otherwise require significant manual analysis.
The implications extend beyond efficiency. A human supervisory team can examine only a finite number of documents, reports, risk indicators and institutions. AI does not remove the need for that expertise, but it can dramatically increase the amount of information that can be screened before human judgement is required.
For banks, that could make the quality of underlying regulatory data increasingly important.
Considerable investment has historically gone into ensuring that the final regulatory submission is accurate. In a more technology-enabled supervisory environment, regulators could become better equipped to examine inconsistencies across datasets, identify unusual movements and compare information across institutions and reporting periods. The emphasis may gradually expand from whether a regulatory report is correct to whether the data infrastructure producing that report is reliable, traceable and internally consistent.
That distinction matters because regulators face many of the same challenges banks encounter when implementing AI. Fragmented data, legacy systems, information security, model accuracy and governance can all determine whether sophisticated analytics produce useful intelligence or simply process poor-quality information more quickly.
Regulatory infrastructure therefore has to evolve alongside the technology.
The BIS Innovation Hub’s Project AISE provides an indication of where this could be heading. The initiative is exploring how central banks and supervisory authorities can responsibly use AI for supervisory analytics, including approaches involving machine-readable reporting, knowledge graphs and model-based analysis. Importantly, the work also considers issues such as unsupported AI outputs, manipulation, reproducibility and accountability rather than treating AI simply as another productivity tool.
These safeguards matter because supervisory AI presents a very different risk profile from many ordinary enterprise applications. A system summarising an internal document can be useful even if an employee occasionally needs to correct its output. A system helping a regulator identify potential weaknesses within a financial institution requires a much higher level of evidence, traceability and control.
Human judgement is therefore unlikely to disappear from supervision. Instead, AI could change where that judgement is applied.
Rather than supervisors manually searching through enormous quantities of information to find potential problems, technology could increasingly perform the first layer of analysis and direct specialists towards areas requiring closer examination. The regulator would still determine whether a signal represents a genuine concern, but the universe of information that can be examined could become substantially larger.
The shift could also move banking supervision closer to continuous awareness rather than periodic snapshots. Advanced analytics could potentially identify changes in liquidity, credit quality, operational incidents, cyber activity or other indicators earlier than conventional reporting processes.
This does not necessarily mean regulators will continuously monitor every transaction occurring within a bank. A more realistic development is that supervisory systems become increasingly capable of detecting changes and anomalies within the information regulators already receive, allowing supervisory attention to be directed more quickly towards areas where risk appears to be increasing.
That could have consequences for compliance technology inside banks as well.
If regulators become better at finding unusual patterns across banking data, financial institutions will have a strong incentive to identify the same patterns internally before their supervisors do. Regulatory technology could therefore evolve from systems primarily designed to prepare submissions and demonstrate compliance towards systems that continuously examine an institution from something resembling a regulator’s perspective.
Banks could increasingly use AI to search for inconsistencies across their own reporting data, identify unusual changes in risk exposures and determine whether emerging patterns could attract regulatory attention. Compliance would become more predictive rather than predominantly retrospective.
This creates an interesting technological dynamic. Banks would use AI to operate and monitor themselves, while supervisors would use AI to independently analyse the information generated by those institutions.
The relationship becomes increasingly machine-assisted on both sides.
There is also a wider supervisory challenge. Regulators are not only responsible for understanding how banks use AI; they must increasingly consider how widespread AI adoption throughout the economy changes the risks banks face.
The technology can increase the speed and sophistication of cyber threats, create dependencies on a relatively small number of technology and cloud providers, and introduce new operational risks when models become embedded within important banking processes. Over a longer horizon, AI could also affect employment, industries and business models, potentially altering the creditworthiness of borrowers and changing the economic environment in which financial institutions operate.
Supervisory infrastructure may therefore need to look beyond conventional banking indicators. Scenario analysis, business-model reviews, technology dependencies and cross-sector intelligence could become increasingly important alongside traditional capital and liquidity measures as regulators attempt to understand risks emerging from technological change elsewhere in the economy.
The challenge becomes even more interesting because regulators are simultaneously establishing expectations for how financial institutions themselves use AI.
Governance, accountability, transparency, explainability, data privacy and continuous monitoring are becoming important principles within emerging financial-sector AI frameworks. Banks are increasingly expected to understand the systems they deploy, document how they operate and retain human responsibility for consequential decisions.
Supervisors adopting AI face many of the same questions.
A regulatory authority cannot assume that an AI-generated warning is correct simply because a sophisticated model produced it. Data quality, model limitations, false positives, explainability and security remain relevant whether the technology is being operated by a commercial bank or the institution supervising it.
The future of financial supervision is therefore unlikely to involve algorithms independently deciding whether a bank is compliant. A more realistic direction is an increasingly sophisticated partnership between technology and supervisory judgement, where machines process a much larger information environment and humans remain responsible for interpreting what those systems discover.
For banks, the implications extend beyond regulatory compliance. Institutions have spent years modernising their technology so they can operate in a faster and increasingly data-driven financial system. They may now need to consider whether their regulatory architecture, data governance and reporting infrastructure are ready for supervisors undergoing the same transformation.
Banks have become accustomed to regulators asking more questions as financial services become more complex. The next change may be that regulators have increasingly powerful technology helping them decide which questions to ask.
What it means
- Banking supervision could become more continuous. AI and advanced analytics may allow regulators to identify emerging patterns between conventional reporting cycles rather than relying exclusively on periodic snapshots.
- Underlying data quality becomes increasingly important. As supervisory technology becomes more capable of comparing and analysing information, inconsistencies within banking data could become easier for regulators to identify.
- RegTech could become more predictive. Banks may increasingly use their own AI and analytics to identify anomalies and potential regulatory concerns before similar signals are detected by supervisors.
- SupTech could significantly expand supervisory capacity. AI can screen far more information and direct regulatory attention towards areas where professional judgement and deeper investigation are required.
- Regulators face their own AI governance challenge. Accuracy, explainability, security, provenance and accountability become particularly important when AI is used by the authority responsible for supervising financial institutions.
- AI is emerging on both sides of the regulatory relationship. Banks will increasingly use AI to operate and manage risk, while regulators use AI to understand and supervise those activities.
