When Banking Goes Offline
What Japan, France, and Britain Reveal About Financial Resilience in the AI Age
The Human Wisdom Project | Applied Human Wisdom
When Money Becomes Inaccessible
On November 6, 2026, a technical failure disrupted banking services at 128 Shinkin financial institutions across Japan.
Customers at affected institutions encountered problems withdrawing cash, transferring money, and accessing online banking. The disruption began shortly after 1:25 p.m. and continued until approximately 7:05 p.m.
An equipment malfunction caused the outage, not a cyberattack.
For nearly six hours, technology that normally allows people to manage their money within seconds became unavailable.
Imagine needing to withdraw cash for an urgent expense or transfer funds to a supplier before the end of the business day.
The money may still be in the account. The bank may remain financially sound. Yet the customer may not be able to use those funds when they need them.
This exposes a distinction that is easy to overlook.
Financial security depends not only on the money people possess but also on their ability to access and use it.
Japan’s disruption came amid heightened concern about the reliability of digital financial infrastructure.
Two days earlier, a French banking group disclosed a major cyberattack involving fraudulent transactions. In October, a Google Cloud disruption had affected online financial services in Britain.
The incidents had different causes. None has been established as an AI-related failure.
Together, however, they reveal vulnerabilities that deserve greater attention as banking becomes more interconnected, automated, and dependent on advanced technology.
What does financial security mean when money is still there, but the systems that make it usable have stopped working?
1. Three Countries, Three Forms of Financial Vulnerability
Modern banking depends on numerous systems functioning together.
Accounts must remain accurate. Payments must be authenticated and processed. Digital applications must communicate with transaction networks, while the infrastructure supporting these services must remain available.
Failures can emerge at different points within this chain.
Recent events in Japan, France, and Britain illustrate three distinct vulnerabilities.
Japan: When a shared system fails
On November 6, a malfunction within the Shinkin Kyodo Center’s shared banking infrastructure disrupted services at 128 Shinkin institutions, mainly in western Japan.
Shinkin banks are cooperative financial institutions that play an important role in supporting local communities and smaller businesses.
Shared technology allows them to provide banking services without each institution having to develop and maintain every system independently.
But it also creates shared exposure.
When equipment within the central system failed, the consequences extended across numerous institutions.
According to the Shinkin Kyodo Center’s restoration announcement, the disruption began at 1:27 p.m., and services were restored by 7:05 p.m.
The operator attributed the incident to equipment failure and reported no evidence of external unauthorized access.
Yet the return of technical services did not mean that every affected transaction was immediately completed.
Kurayoshi Shinkin Bank announced that certain transfers to other financial institutions delayed by the disruption would be credited on the next business day, November 9.
The system was restored on Friday. Some delayed payments would not reach their destinations until Monday.
This difference between technical restoration and the completion of customer transactions will become central to understanding financial resilience.
France: When financial security is breached
On November 4, the French banking group Crédit Mutuel Arkéa disclosed a sophisticated cyberattack that enabled fraudulent transactions involving a bank account.
Reporting on November 5 placed the amount stolen at approximately €100 million, although the final net loss remained uncertain as recovery efforts continued.
The bank temporarily suspended certain transactions to contain the incident and strengthen security.
Crucially, it said customer assets were not affected.
In a subsequent explanation, Arkéa said the attackers had reached the bank through a third-party service provider’s platform, not through customer accounts or its online subsidiary Fortuneo.
The incident exposed a different vulnerability.
A financial institution can suffer a serious security breach through a system it does not directly operate. Even when customer funds remain protected, the measures needed to contain an attack can delay legitimate transactions.
The French case was not a nationwide banking collapse. It was a significant security incident affecting one banking group.
Its importance lies in the relationship between institutional security, external technology providers, and service continuity.
Britain: When cloud infrastructure becomes unavailable
On October 20, an incident involving Google Cloud’s London region disrupted numerous digital services.
Online services associated with British financial institutions, including Lloyds, Halifax, HSBC, and Revolut, experienced difficulties during the disruption.
Google’s incident report, published on November 4, traced the problem to an external electricity supply failure at a London data center.
The consequences were intensified by weaknesses in backup power controls.
During maintenance, the facility was operating with reduced electrical redundancy. When the active power supply failed, a configuration problem prevented backup generators from starting automatically.
Battery reserves were eventually exhausted, interrupting computing operations.
As workloads shifted to other infrastructure, delays increased, and error rates rose.
The principal service impact was resolved within approximately six hours.
This was not a banking cyberattack or an AI malfunction.
It was a physical infrastructure failure whose effects reached digital services used by financial institutions.
The three incidents therefore involved different mechanisms: equipment malfunction, malicious intrusion, and external infrastructure disruption.
Yet their consequences converged.
Whether the failure begins inside a bank, through an external provider, or within a shared technology system, customers may lose access to essential financial services.
Understanding why these disruptions can spread requires examining the infrastructure behind modern banking.
2. The Hidden Fragility of Shared Digital Infrastructure
Modern financial services increasingly depend on infrastructure shared across institutions.
Banks use common payment-processing systems, cloud platforms, telecommunications networks, and specialized technology services rather than maintaining every function independently.
This arrangement offers substantial advantages.
Shared processing reduces costs and improves standardization. Cloud services provide scalable computing capacity and sophisticated technical capabilities. Smaller institutions can access infrastructure that would otherwise be difficult to maintain.
But shared efficiency can also create shared vulnerability.
When several institutions depend on the same provider or technical component, they may face the same disruption.
This is known as concentration risk.
A related concern is a common point of failure: a system or component whose malfunction can interrupt multiple services or organizations.
A bank may appear operationally independent while relying on technology also used by dozens of other institutions.
The Japanese incident illustrates the consequences of a shared system malfunction. The French attack reveals how an external provider can become a route into a financial institution. The British outage shows how physical infrastructure failures can affect services operating far beyond the data center itself.
These cases do not prove that shared infrastructure is inherently unsafe.
A well-managed shared system may provide greater security and reliability than numerous independently maintained systems.
The risk emerges when institutions underestimate the consequences of losing a critical dependency.
When backup systems share the same weakness
Backup arrangements are intended to preserve operations during disruptions.
But redundancy works only when alternative systems can function under the same conditions that disable the primary system.
Google’s October incident illustrates the problem.
Backup power equipment existed, but a configuration weakness prevented the generators from starting as intended. Systems designed to provide protection were affected by a common operational condition.
This is an example of common-mode failure: a situation in which supposedly separate safeguards become ineffective because they share an underlying vulnerability.
The lesson extends beyond electricity.
A financial institution may have multiple applications, recovery facilities, or service providers while still depending on a single critical component.
The number of backup arrangements matters less than their ability to operate independently when needed.
Institutions must therefore understand not only which technologies they use but also how those technologies depend on one another.
The vulnerability of shared infrastructure lies not simply in interconnection but in failures that spread across connections institutions have not adequately understood or tested.
This structural challenge becomes more complex as AI is integrated into banking operations.
3. What Changes When Banking Becomes AI-Driven?
Artificial intelligence introduces a different kind of uncertainty into financial services.
The incidents in Japan, France, and Britain were not established as AI failures. But they show why technological complexity and institutional control matter.
As banks adopt AI, they must manage not only whether systems remain operational but also whether automated outputs and decisions remain accurate, explainable, and correctable.
AI can improve customer service, document analysis, fraud detection, risk monitoring, and internal operations.
These benefits are already encouraging widespread adoption.
According to the Bank of Japan’s August 24, 2026 report, more than 90% of the 150 financial institutions surveyed were using or trialing generative AI.
Use cases were expanding beyond administrative assistance toward more central business functions involving customer information.
Yet the Bank of Japan also identified continuing challenges in governance, information management, safety, security, and the uncertainty of AI-generated outputs and behavior.
When plausible answers are not reliable answers
Conventional software can contain defects, but its intended behavior is generally governed by defined instructions.
Generative AI introduces additional uncertainty.
Its outputs may sound authoritative but be inaccurate. Responses can change with inputs, context, model updates, and data quality.
In banking, this matters when AI-generated information influences customer communication or institutional decisions.
An incorrect summary may distort a staff member’s assessment. An automated assistant may provide incomplete guidance. A model may perform well in ordinary situations but respond poorly to an unfamiliar case.
Not every AI error produces financial harm.
The challenge is determining which functions can safely tolerate uncertainty and which require stronger safeguards.
A system used to draft internal notes does not necessarily require the same controls as one influencing a financial decision.
Risk management must reflect those differences.
When AI moves from advice to action
AI becomes more consequential when it can initiate actions rather than merely provide information.
An assistant that summarizes a document presents a different level of risk from an automated agent authorized to update records, initiate workflows, or trigger transactions.
As authority increases, so does the need for clear limits.
Who approves the action? What information can the system access? When must a human intervene? Can an incorrect decision be reversed?
Human oversight must be practical.
Assigning responsibility to an employee is insufficient if that person lacks the information, authority, or time required to intervene effectively.
Institutions must ensure that automated decisions remain within defined boundaries and that exceptions can be identified promptly.
Governing AI as part of financial operations
AI governance cannot stop at testing whether a model produces useful answers.
Financial institutions must understand how models are used, what information enters them, how outputs influence decisions, and what happens when performance changes.
Externally supplied models also require appropriate validation and oversight. The issue here is not simply reliance on another technology provider, but whether the institution can control the decisions and actions supported by the model.
The Bank of Japan’s October 2026 Financial System Report identifies advances in frontier AI as a cybersecurity concern, even as AI may strengthen defensive capabilities.
These developments do not mean that AI inevitably weakens banking security.
Well-governed AI can improve monitoring, detect unusual activity, and support faster responses.
The important question is whether the institution’s ability to govern the technology develops alongside its capabilities.
The greatest uncertainty in AI-driven banking is not intelligence itself, but whether increasingly autonomous systems remain understandable, controllable, and accountable.
That leads to a distinction which applies to banking technology regardless of whether AI is involved.
4. When Efficiency Is Not the Same as Resilience
Digital banking has transformed customer expectations.
People expect instant transfers, continuous access, low transaction costs, and services available without visiting a branch.
Technology helps financial institutions meet those expectations.
Get Nattakitt Udomphol’s stories in your inbox
Join Medium for free to get updates from this writer.
But a system designed for efficient operation is not necessarily prepared for disruption.
Three concepts need to be distinguished.
Efficiency concerns how effectively resources deliver services.
Reliability concerns whether those services function correctly and consistently under expected conditions.
Resilience concerns whether essential services can withstand disruption, limit its consequences, and recover safely.
A system may be efficient and reliable during normal operations while remaining vulnerable to unusual combinations of failures.
Resilience therefore requires capabilities whose value may be difficult to see when everything works.
Alternative service channels, contingency procedures, realistic testing, and recovery arrangements all involve costs.
Yet their purpose is different from improving daily transaction speed.
They protect essential functions when normal operations become impossible.
Not every financial service requires the same level of protection.
A temporary interruption to a nonessential reporting tool is different from an inability to withdraw money or complete an urgent payment.
Resilience must reflect the consequences of interruption, not merely the technical importance assigned to a system.
System recovery is not customer recovery
The Japanese outage offers a particularly revealing example.
The shared banking system resumed operation on November 6. Yet certain delayed transfers at Kurayoshi Shinkin Bank were scheduled to reach recipient accounts on November 9.
Technical restoration and the completion of affected customer transactions occurred on different timelines.
This reveals two dimensions of recovery.
System recovery occurs when the technological infrastructure required to deliver a service resumes operation.
Customer recovery occurs when the practical consequences of disruption have been sufficiently resolved for customers to resume essential financial activities.
Customer recovery may involve completing delayed transfers, reconciling transactions, confirming account information, and resolving outstanding inquiries.
These processes do not necessarily finish when servers restart.
Nor does every delayed transaction cause financial harm. Its consequences depend on timing and individual circumstances.
Nevertheless, measuring recovery only through technical availability can overlook what customers continue to experience.
A bank’s technology team may consider a system restored when it can process transactions again.
A business waiting for a delayed payment may reach a different conclusion.
Both perspectives matter, but they answer different questions.
This distinction has implications for institutional reporting and accountability.
Recovery measures should examine not only when technology becomes operational, but also whether outstanding customer transactions and essential services have been restored.
From the customer's perspective, a financial system is not fully restored merely because its technology is running again.
That principle provides a foundation for evaluating what institutions should be expected to prepare, demonstrate, and improve.
5. What Should Financial Institutions Guarantee?
No financial institution can promise uninterrupted technological operation under every circumstance.
Equipment can malfunction. Software contains defects. Security threats evolve. External infrastructure may become unavailable.
A realistic responsibility is to identify essential services, prepare for disruption, and demonstrate that recovery arrangements work.
Regulators increasingly recognize operational resilience as a central obligation.
In the United Kingdom, the Financial Conduct Authority requires firms in its operational-resilience regime to identify important business services, set disruption tolerances, test vulnerabilities, and strengthen recovery capabilities.
In-scope firms were required to demonstrate their ability to operate within those tolerances by March 31, 2025.
The European Union’s Digital Operational Resilience Act (DORA), applicable since January 17, 2025, establishes requirements covering ICT risk management, incidents, resilience testing, and technology-related third-party risks.
These frameworks operate under different legal arrangements. Neither guarantees that disruptions will disappear.
They do, however, establish that resilience requires more than technical confidence or written contingency plans.
Building on these principles, five questions can help assess whether financial institutions are prepared to protect essential services.
The Five-Part Financial Resilience Test
The following test is the author’s analytical proposal, not an official regulatory methodology.
1. Essential Service Continuity
Can the institution identify which services must remain available — or be restored within defined limits — during disruption?
Not all financial services carry the same consequences when interrupted.
Access to funds, time-sensitive payments, accurate transaction information, and essential customer support may require stronger safeguards than noncritical administrative functions.
Institutions should identify priorities, establish realistic disruption limits, and provide alternative channels where appropriate.
The test is whether those arrangements reflect the consequences for customers, not merely internal technical classifications.
2. Dependency Transparency and Testing
Can the institution demonstrate that it understands its critical dependencies and has tested how they behave under failure?
This requires identifying the systems, providers, and connections supporting essential services.
Testing should include severe but plausible scenarios, including simultaneous disruptions and unsuccessful transitions to backup arrangements.
Evidence should extend beyond the existence of contingency documents.
It should include test results, weaknesses identified, corrective actions, and subsequent verification.
The objective is to establish whether the institution’s arrangements work under realistic pressure.
3. Human Oversight of Automated Decisions
Can important automated decisions be monitored, explained, interrupted, and corrected?
As AI moves into financial operations, institutions need clear boundaries governing the authority delegated to automated systems.
Appropriate controls may include approval requirements, audit trails, escalation procedures, and error-correction mechanisms.
Oversight must be meaningful.
Employees responsible for supervising high-impact automated processes need sufficient information and authority to intervene.
This criterion examines whether automation remains governable when unusual conditions arise.
4. Customer-Centered Recovery
Can the institution show that customers have regained effective access to essential services after disruption?
Recovery assessments should extend beyond the moment when technology becomes operational.
Relevant evidence may include unresolved transaction backlogs, payment-reconciliation progress, restoration of customer access, and complaint resolution.
Customers should receive timely and understandable information about service interruptions and pending transactions.
When consequential harm occurs, accessible procedures should exist to assess and remedy it under applicable law and contractual obligations.
This does not require automatic compensation for every inconvenience.
It requires institutions to address the practical consequences of disruption rather than treating technical restoration as the end of their responsibility.
5. Accountability and Institutional Learning
Does each significant incident produce evidence that the institution has reduced the likelihood or consequences of recurrence?
Investigations should examine not only the immediate cause, but also weaknesses in governance, testing, communication, and operational procedures.
Boards and senior executives should have visibility over material risks and unresolved corrective actions.
Lessons should lead to identifiable improvements, with clear responsibilities and verification that changes have been implemented.
Repeated weaknesses should not be treated as unrelated surprises.
From compliance to demonstrated resilience
The value of these five questions lies in the evidence they require.
A policy document does not prove that essential services will remain accessible.
A recovery plan does not prove that recovery will succeed.
An AI governance policy does not prove that automated decisions remain controllable.
Testing, monitoring, and operational experience must demonstrate those claims.
Institutions should therefore evaluate both technical performance and customer outcomes.
How long were essential services unavailable? Did backup arrangements operate as expected? Were delayed transactions completed accurately? Did customers receive clear information? Were identified weaknesses corrected?
Answers will differ by an institution’s size, activities, and importance within the financial system.
Resilience does not require identical technical arrangements for every bank.
It requires protection proportionate to the consequences of failure, supported by verifiable evidence.
Financial institutions are not responsible for promising that technology will never fail. It is to show that essential services, recovery arrangements, and accountability are designed for the possibility that it will.
Conclusion: Trust Requires Systems That Can Fail Safely
Digital banking has transformed everyday economic life.
It allows people to transfer money instantly, manage accounts remotely, and use financial services that once required considerable time and effort.
Cloud infrastructure and AI may extend these benefits, making banking more accessible, responsive, and efficient.
But greater convenience also creates deeper dependence.
When financial services become integral to daily life, a technical disruption can quickly become a practical problem for households and businesses.
The experiences of Japan, France, and Britain show that such failures have no single cause.
They can emerge from ordinary equipment malfunctions, malicious attacks, or problems within external infrastructure.
AI introduces additional uncertainty as institutions automate more decisions and rely on increasingly complex technology.
Yet the underlying responsibility is not new.
Financial institutions must safeguard assets and maintain financial soundness. They must also consider the continuity and accessibility of the services people depend on.
An account balance offers limited reassurance when funds cannot be accessed at a critical moment.
A recovery announcement may mean little to someone whose payment remains unresolved.
And an intelligent automated system cannot inspire lasting confidence if its decisions cannot be understood, corrected, or challenged.
Financial resilience is therefore more than a technical property.
It reflects an institution’s capacity to anticipate disruption, preserve essential functions, respond responsibly, and restore services with attention to their human consequences.
The future of banking will become more digital, more interconnected, and increasingly intelligent.
But greater intelligence does not automatically produce greater trust.
That trust must be earned through systems designed to withstand disruption and institutions prepared to remain accountable when those systems fail.
In an increasingly intelligent financial system, the greatest test of trust may not be what technology can do when everything works — but how institutions protect people when it does not.
Sources and Further Reading
Japan: Banking Operations and AI Governance
- Shinkin Kyodo Center. (2026, November 6). System restoration and resumption of banking services.
- Kurayoshi Shinkin Bank. (2026, November 6). Notice of recovery from the online banking system disruption.
- Bank of Japan. (2026, August 24). Use and Risk Management of Generative AI by Japanese Financial Institutions: Based on the Results of FY2026 Survey.
- Bank of Japan. (2026, October 23). Financial System Report.
France: Cybersecurity and Financial Operations
- Crédit Mutuel Arkéa. (2026, November 4). Crédit Mutuel Arkéa announces that it has been the victim of an attack.
- Crédit Mutuel Arkéa. (2026, November 4). Additional information following the attack.
- Reuters. (2026, November 5). French bank Crédit Mutuel Arkéa says €100 million missing after cyberattack.
Britain: Cloud Infrastructure and Operational Resilience
- Google Cloud. (2026, November 4). Incident report: Multiple services experiencing issues in the europe-west2 region.
- Associated Press. (2026, October 20). Google Cloud outage hits UK banks.
- Financial Conduct Authority. Operational resilience: Requirements and guidance for financial firms.
European Financial Regulation
The three incidents involved different causes, and no evidence presented establishes that AI directly caused any of them. Financial-loss estimates and incident findings may be updated. The Five-Part Financial Resilience Test is the author’s analytical framework, not an official regulatory standard. Illustrative customer consequences are not claims of documented individual losses.
