Japanese Banks Warn AI Cyberattacks Could Disrupt ATMs and Online Banking

A banking crisis once looked like a line outside a branch: depositors holding passbooks, waiting to withdraw cash. In 2026, a crisis may begin more quietly. A phone screen says the service is unavailable. An elderly customer pauses in front of an ATM. A company payment is delayed. Markets are open, but settlement has slowed. The modern bank crisis can happen not at the vault, but in the code and connections that make money move.

In June, the head of Japan’s banking lobby warned that lenders may have to suspend services such as ATMs and online banking if sophisticated AI-enabled cyberattacks become a serious threat to the banking system. Reuters reported that concern has intensified around frontier AI systems, including Anthropic’s Mythos, which are feared to be capable of identifying software vulnerabilities at unusual speed.

This is not simply a story about weak banks. It is a story about banks becoming the nervous system of society. ATMs, online banking, smartphone apps, direct debits, card payments, salary transfers, corporate payments, securities transactions, BOJ systems, the Zengin network, cloud vendors, outsourced IT providers, APIs and fintech connections all sit inside the same trust architecture. Japan still remembers cash, but its financial life now depends on invisible digital rails.

Why AI changes the threat

AI can help defenders and attackers. Banks already use AI for fraud detection, anti-money-laundering monitoring, customer service, log analysis and cyber threat intelligence. Attackers can use it too: to write natural Japanese phishing messages, tailor scams to specific employees, create fake voices or videos, research targets and prioritize weak points.

Traditional cyberattacks required time, expertise and tooling. AI can automate parts of that work. Criminals who lack elite skills may still generate convincing messages, fake internal documents, social-engineering scripts and reconnaissance. The danger is not only that attacks become smarter. It is that more people can attempt them, faster.

For banks, the problem is not only fraudulent transfers. If AI helps attackers discover many vulnerabilities quickly and launch pressure across multiple systems, banks cannot rely on a slow detect-and-repair cycle. They may need to isolate systems or suspend parts of service before damage spreads. That is why ATMs and online banking enter the discussion.

The FSA and BOJ move to short-term measures

Japanese authorities are treating this as a present risk. On June 15, 2026, the Financial Services Agency published an English translation of a request made jointly with the Bank of Japan regarding “Short-Term Measures for Financial Institutions in Response to Changes in Threat Posed by Frontier AI.” The request had been issued on May 22 and assumes that AI progress is changing the cyber-risk environment for financial institutions.

In May, the FSA also created a practitioner-level working group under the public-private coordination framework on cyber risk in finance. Its stated purpose is to help financial institutions, IT service providers, public bodies, the government and the Bank of Japan share an understanding of threats arising from advances in AI technology and consider responses. Details are not disclosed because they involve cybersecurity information.

In April, Reuters reported that Finance Minister Satsuki Katayama described the issue as an immediate crisis during a meeting involving the FSA, the Bank of Japan, the national cybersecurity office, major banks and Japan Exchange Group. Japanese regulators are not treating AI cyber risk as a distant abstract problem. They are treating it as a financial-stability issue.

Financial cyber risk in numbers

Japan already knows what “banking stopped” feels like

Long before AI, Japan experienced banking systems that stopped. Mizuho is the most famous example. In 2021, Mizuho Bank suffered repeated system problems affecting ATMs and online services. Reuters reported that one March 2021 disruption affected 4,318 of 5,395 ATMs nationwide. The cause was not a cyberattack. For customers, however, the experience was similar: cash did not come out, transfers could not proceed and cards were trapped or delayed.

In October 2023, Japan’s Zengin interbank data communication system suffered a major outage. Reports said 11 banks were affected and more than 1.4 million transactions were disrupted. Zengin is a core part of Japan’s interbank transfer system, dating back to the 1970s. When it fails, the effects go beyond individual customers. Corporate payments, supplier transfers, salaries and settlement routines can all be affected.

These incidents were not AI attacks. That is precisely why they matter. Japan already knows how society reacts when financial infrastructure stops. AI adds another layer: attack speed, attacker scale and the possibility of simultaneous pressure across interconnected systems.

The ATM as Japanese infrastructure

In Japan, an ATM is not just a machine. It is a social utility. Convenience stores, post offices, train stations, bank branches and supermarkets all use ATMs to connect people to cash. Elderly customers, rural residents, small merchants, cash businesses and foreign visitors all depend on them. Japan is moving toward cashless payments, but trust in cash remains strong. ATM disruption is felt as an infrastructure problem.

Online banking is just as critical. Younger customers check balances, invest, pay rent, receive salaries and pay taxes on their phones. Companies manage cash flow and pay suppliers online. As branches shrink and digital banking grows, online-service interruption becomes more important.

That means AI cyber resilience is not only a bank problem. It involves convenience-store ATM networks, telecom carriers, cloud providers, card companies, payment processors, local governments, police, the Bank of Japan, the FSA, corporate treasury departments and customer behavior.

What defense means in the frontier-AI era

The first defensive priority is vulnerability management. Banks need to know which systems, vendors, APIs, cloud services, outsourced functions and third-party products they rely on, and what vulnerabilities matter most. In the AI era, attackers may do that mapping quickly; defenders must be faster.

The second priority is designed shutdown and recovery. Not every service needs to stop at once. A bank may need to limit ATM functions, lower transfer limits, isolate corporate payments, restrict app features or delay certain transactions. These choices must be designed before a crisis. If they are improvised during an attack, they will be slow and confusing.

The third priority is communication. If a bank stops services suddenly, customers panic. Rumors spread on social media. Criminals send fake “recovery” emails and SMS messages. Banks must prepare official channels, customer notices, fraud warnings, recovery timelines and alternative-service instructions before the incident happens.

The difficulty of using AI to defend AI-era finance

Banks will use AI for defense: anomaly detection, fraud patterns, threat intelligence, fake-site detection, customer authentication and case summaries. But defensive AI has its own risks. Too many false positives block legitimate customers. Missed signals allow fraud. If the defensive model itself is attacked or poorly governed, it becomes part of the risk.

A 2026 finance-focused study on AI-driven cyber threat intelligence found that effective use depends on governance, integration into security workflows, analyst trust and security for the AI models themselves. In other words, buying an AI tool is not enough. Banks need evidence, monitoring, auditability, human responsibility and operational integration.

The contest between attack AI and defense AI is not only a model-performance race. It is an organizational race. Who sees the logs? Who decides? Who has authority to stop a service? Who informs customers? Who takes responsibility? In AI-era banking, resilience depends on decision speed as much as technology.

What customers should know

For ordinary customers, the answer is not fear but preparation. Use only official bank apps and official websites. Do not log in from links in SMS or email messages. Enable two-factor or biometric authentication. Check payee information carefully before large transfers. Keep a small amount of cash and more than one payment method in case one channel is temporarily unavailable.

The most dangerous moment may be during a service outage. Scammers know that confused customers are more likely to click fake “recovery” links. If banking services are disrupted, return to verified channels: the official app, the official website typed directly into the browser, the number on the back of a bank card, or physical branch notices.

Japan.co.jp’s view

The banking lobby’s warning is not weakness. It is a sign that banks are beginning to speak honestly about AI-era risk. The best financial infrastructure does not stop. But when an attack is spreading, a controlled stop may be the only way to preserve trust.

Japan is a country that still trusts cash while increasingly depending on digital payment rails. Protecting both requires banks, the FSA, the BOJ, IT providers, telecom companies, police and customers to share the same sense of urgency. AI does not only make banks more vulnerable; used properly, it can make defense faster. But a bank that simply trusts AI is not resilient. A bank that understands AI, can override it, can stop services safely and can explain itself is stronger.

Financial trust is not made only of interest rates and capital ratios. It is made of ATMs that work, salaries that arrive, companies that can pay suppliers, app balances that are correct, and explanations that arrive before panic does. In 2026, the question for banks is not only how convenient they can be. It is whether they can preserve trust when convenience has to pause.

Reader guide