Press review: la Repubblica, 24 October 2026. With the comment of attorney Fulvio Sarzana, Sarzana & Partners Law Firm.
In short. In the United States, banks report AI agents that enter users’ current accounts and in some cases carry out unauthorised transactions. According to attorney Fulvio Sarzana, consulted by Repubblica, Italian sector rules do not cover the case and nothing obliges the bank to refund the customer. Recourse against the software producer may be possible, but it is a route still to be tested.
The daily la Repubblica, one of Italy’s main newspapers, published on 24 October 2026 an article by Alessandro Longo titled «Banks, AI enters the account. The alarm in the United States» (in Italian, subscribers only). This is a summary of its main points on AI agents, with the legal part in which Fulvio Sarzana is quoted.
What is happening in the United States
The reports come from the lending institutions themselves: operations carried out autonomously by agents through online banking. US banks see AI agents entering users’ current accounts and, in some cases, taking unauthorised actions.
According to Repubblica, some institutions have begun to block AI agents: Morgan Stanley, Bank of America, JPMorgan Chase and Citigroup.
Not just hacker attacks
It is not always a matter of attacks by programs out of control. A growing number of people in the United States entrust AI agents with checking the latest transactions and emailing a periodic report. In some cases the agent misreads the instructions and makes transfers to third parties.
AI agents are coming to Italy too
The issue concerns Italy as well. Dario Fadda, an information security analyst for a large Italian bank, explains that Italian account holders could also use AI agents for automatic access to their accounts. For banks it is very hard to tell them apart from ordinary people using e-banking.
If it has not yet happened on a large scale, the article notes, it may be because the best-known AI agent, Meta’s Muse, is not yet available in Europe. It was launched in September and is already used by about 10 million people.
Systemic risks for the economy
AI agents can be asked to handle tasks such as switching energy supplier or bank in search of better contracts. The banking system is not designed for an army of agents moving current accounts and liquidity en masse, at superhuman speed.
Experts and institutions are already discussing it. In May the European Central Bank published a study simulating a similar scenario, with devastating effects on financial stability, similar to those of a mass withdrawal of cash from current accounts.
AI agents: who is liable for the agent’s actions
If the agent causes a financial loss, who is liable? According to attorney Fulvio Sarzana, current sector rules say nothing on this topic and nothing obliges the bank to refund the customer. The exceptions are cases where the unauthorised access is the work of criminals or a fraud.
Sarzana adds that it may be possible to claim against the software producer for its alleged defects, but this is a route still wholly untested.
The general framework for payments is Legislative Decree 11/2010, which transposes the EU payment services directive. It provides for refunding the customer for unauthorised transactions, except in cases of fraud, intent or gross negligence by the customer. Whether a transfer ordered by an AI agent chosen by the customer is «unauthorised» remains an open question.
What to do in the meantime
Pending rules, some practical precautions for anyone who uses or plans to use AI agents with their own account:
- read the online banking contract on credentials and use by third parties;
- avoid handing online banking credentials to an AI agent;
- limit the agent to read-only functions, where the service allows it;
- keep the instructions given to the agent and the reports received;
- dispute any suspicious transaction with the bank immediately.
On the topic, see also the firm’s pieces on bank scams and on the AI Act and GDPR (both in Italian).
Frequently asked questions on AI agents and bank accounts
They are artificial intelligence systems to which the user entrusts tasks such as checking the latest transactions, emailing a periodic report or switching energy supplier or bank. They operate through online banking.
Banks report AI agents that enter users’ accounts and in some cases carry out unauthorised transactions. According to Repubblica, Morgan Stanley, Bank of America, JPMorgan Chase and Citigroup have begun to block them.
According to Dario Fadda, an information security analyst for a large Italian bank, Italian account holders could also use them, and it is very hard for banks to tell them apart from people using e-banking. Meta’s Muse, the best-known agent, is not yet available in Europe.
According to attorney Fulvio Sarzana, current sector rules say nothing on this topic. It may be possible to claim against the software producer for its alleged defects, but this is a route still wholly untested.
For Sarzana nothing obliges the bank to refund, unlike cases where the unauthorised access is the work of criminals or a fraud. Legislative Decree 11/2010 provides for refunds of unauthorised transactions, and whether a transaction ordered by an agent chosen by the customer qualifies remains an open question.
According to the article, yes: the banking system is not designed for an army of agents moving accounts and liquidity en masse at superhuman speed. In May the European Central Bank published a study simulating a similar scenario, with devastating effects on financial stability.
The full article is in la Repubblica (in Italian, subscribers only).
Italian 
English 