A current account may be temporarily inaccessible via the bank’s app or website without a euro having been stolen. This would happen, for example, if a cyber attack blocked the programs used to consult the balance or make payments. Artificial intelligence, increasingly used by banks and potentially also exploitable by cyber criminals, is therefore at the center of checks by supervisory authorities.
In the United States, controls mainly concern AI agents, programs capable of acting with a certain autonomy. In the euro area, however, the European Central Bank (ECB) has asked the major institutions under its supervision to present a plan by 31 October 2026 to strengthen defenses against cyber attacks that exploit this technology.
How AI is used in banks and what the risks are
Banks already use artificial intelligence to detect possible fraud, analyze data and assist customers. The technologies used also include so-called agentic AI. Unlike a chatbot that answers a question, an AI agent can organize a sequence of actions to achieve a goal, using the tools it has access to. In cybersecurity, for example, it could examine reports from a monitoring program, identify the most urgent ones and, if authorized, initiate a response procedure. This does not mean that you can freely transfer money from your accounts: the operations allowed depend on the permissions received.
The problem arises when an agent misinterprets information or takes an unexpected action. If it mistook legitimate activity for a threat, it could help block a service. However, if it did not recognize an intrusion, it would risk delaying the intervention of the technicians. There is also the possibility that malicious instructions hidden in a document or message are considered trusted commands. It is a technique known as prompt injectionwhich can cause a program to behave differently than expected.
Hackers can also exploit AI to identify vulnerabilities in software and automate certain stages of intrusions. A case documented by Sophos using 12 AI agents shows, for example, how these tools can be used to develop and test programs for cyber attacks, without eliminating human control. The ECB warns that these capabilities can increase the speed and scope of attacks, reducing the time banks have to react.
For account holders, the consequences would depend on the type of incident. A failure could delay wire transfers or other payments, while a records breach would risk exposing confidential data. If a technology provider used by multiple institutions were affected, the difficulties could affect numerous customers at the same time.
What are the security measures
To reduce risks, it is necessary to establish exactly which activities AI agents can carry out and provide human confirmation for the most delicate operations, not to mention that automatic actions must be traceable, with the possibility of quickly blocking the program if it behaves abnormally.
However, protection does not only concern autonomous tools. Institutions must identify software weaknesses, install security updates and monitor suspicious access. Particular attention should be paid to external suppliers, on whom essential functions such as payment processing may depend.
It is equally important to be able to restore operations after an attack. If the main infrastructure is compromised, for example by ransomware, it is not enough to keep a copy of the data: it must be able to recover it in a secure environment and reactivate essential functions. This requires isolated backups from affected networks, emergency procedures and periodic tests to verify whether the recovery plan really works.
In its letter of 7 July 2026, the ECB also identified as a priority the upgrading or replacement of outdated systems, which may be more difficult to keep secure. It also asked the major supervised institutions to specify the resources to be allocated to the interventions, who will have to carry them out and in what timeframes.
From 17 January 2025, the Digital Operational Resilience Act (DORA) regulation applies in the European Union, which establishes obligations regarding IT risk management, incident reporting, resilience testing and control of services entrusted to third parties. It is not exclusively about AI, but aims to prepare financial operators to deal with digital failures and attacks.
How to protect your bank account from AI
The precautions available to account holders serve above all to combat fraud. They cannot prevent a breach of the banking infrastructure, the protection of which is the institution’s responsibility, but they help reduce the risk of unauthorized access to the account.
It is advisable to activate transaction notifications, as well as regularly check movements, so as to quickly identify any unknown charges. It is important to protect access to home banking, do not share passwords or temporary codes and do not approve authentication requests for payments that have never been initiated. Before making a transfer, it is then advisable to check that the name of the beneficiary corresponds to the IBAN, paying attention to any notices from the bank. Even if an AI assistant promises to simplify financial management, your credentials should not be entered into external services not authorized by your bank.
Another precaution concerns communications apparently coming from the credit institution. Scammers can leverage AI to create believable messages and imitate other people’s voices. If a phone call reports alleged anomalies and asks to transfer money, provide codes or share the screen, it is advisable to interrupt the conversation and check directly through the app or the bank’s official number. The fake video calls from so-called banking operators show how a scam can succeed even without violating the institute’s IT systems.
In the event of an unauthorized payment, the bank must be notified and any compromised instruments must be blocked. The dispute must be submitted without delay and, as a rule, no later than 13 months after the charge. The legislation generally provides for reimbursement by the end of the business day following the day on which the intermediary becomes aware of the transaction. The bank may, however, suspend it in the event of justified suspicion of fraud on the part of the customer, which must be communicated immediately in writing to the Bank of Italy. Furthermore, if willful or grossly negligent behavior is demonstrated, the intermediary can deny the refund or recover the sums already returned.
The case of a bank transfer arranged personally after having been deceived is different: in this situation the same protection does not automatically apply. If the dispute is not resolved, it is possible to file a complaint and, when the conditions are met, contact the Financial Banking Arbitrator.
If, however, the problem is a general interruption of home banking, it is best to consult the official notices, check any alternative channels for urgent operations and keep the evidence of the outage.
