How your voice for crimes can be imitated. Artificial intelligence makes the bank fraud difficult to recognize

The bank frauds no longer resemble the ten years ago. Artificial intelligence already allows the perfect imitation of a customer's voice, and this endangers the traditional security methods used by some banking institutions.

Photo source: Getty Images

Sam Altman, the Openai CEO, warns that banks risk a massive security crisis if they do not urgently give up vocal authentication. In an intervention at the conference organized by the US Federal Reserve, Altman stressed that artificial intelligence is already able to perfectly imitate the voice of a person to move on to identity checks and to evade funds from bank accounts.

“One thing that frightens me is that there are still financial institutions that accept the vocal imprint as an authentication method. It is crazy. AI has completely exceeded this technology.”Altman said, for Associated Press (AP).

Voice authentication has become popular among bank customers with big fortunes for a decade. Customers were asked to speak a special phrase on the phone to have access to their accounts, but this method is now vulnerable to AI's vocal clone, Altman warns. “Vocal clones and soon video can imitate people in an increasingly difficult to reality”, emphasized the Openai CEO.

The discussion was moderated by Michelle Bowman, vice -president responsible for supervision within the federal reserve and the main financial regulator of the institution. This suggested that a collaboration between bank authorities and tech companies could be essential for the development of new verification methods: “Maybe it would be something we can reflect on a partnership,” Bowman said.

Against this alert, the pressure increases on banks to give up old authentication methods and invest in more robust solutions, such as multifactorial authentication, hardware tokens and advanced cryptography.

“The good training beats the bad AI”

“AI increased the risk of phishing attacks, but also the level of preparation against them. Our studies show that a good training can overcome a bad.”recently said Pyry Åvist, CTO and co -founder Hoxhunt. In turn, Mika Aalto, CEO of Hoxhuntl, said: “In the near future, you will supply more and more attacks: from text messages, to video and audio deepfakes. Everything will become cheaper, more convincing and easier to use by attackers.”

This “refinement” of the attacks increases their success rate considerably, and the only real protection measure remains aware: the user education to recognize the traps, to check the source and to react with caution

A recent study by accenture shows that 80% of cyber security leaders in the banking sector believe that attackers already have an advantage due to AI.

In this context, financial institutions, whether they are classic or digital banks, are forced to quickly rethink their authentication systems.

Part of the non-Banns, such as Revolut, have already strengthened their infrastructure with artificial intelligence-based solutions and proactive functionalities such as apps or biometric authentication. In parallel, traditional banks also work on adapting systems, but, having larger networks and more complex structures, implementations may involve different stages

Revolut: 64% of Romanian frauds start from the meta platforms

“Frauds are a serious challenge for the entire financial-banking industry, and we carefully treat this phenomenon, managing to prevent losses worth over 700 million euros in 2024, among our own customers. We make every effort to be one step ahead of the organized crime, which makes rapid progress as technologies are becoming more and more accessible.” Revolut representatives told the truth.

In order to increase the safety of customer funds, the institution has launched new security features, such as apps in the application, which help customers avoid SCAMs by impersonation, but also artificial intelligence systems to detect real-time fraud. Along with other measures such as transactions limits and biometric authentication,

“In general, as we added new instruments we saw how they contributed to the increase of the fraud identification and allowed us to increase the total amount of the protected funds. Thus, in Romania, in the last six months (January-June 2025) the total amount of the preventive fraud was 32% higher than the total amount that we succeeded in a total It was maintained as a whole at the same level ”they said.

In fact, the company recently published a new report on the protection of customer funds and fraud prevention. The document describes an accelerated evolution of tactics used by people who commit fraud, both globally and locally, and emphasizes that encrypted messenger platforms become more and more a starting point for attacks.

According to the report, in the second half of 2024, the largest increases in the reported fraud were associated with WhatsApp and Telegram. Globally, 39% of the reported frauds were propagated by these platforms, and in Romania the percentage was 40%. At the same time, the meta platforms (Facebook, Instagram, WhatsApp) were involved in over 54% of the cases reported globally and in 64% of those in Romania. This trend has been constantly maintained in all three reporting periods so far, according to the quoted source.

The document also shows that the most common frauds are those of “online shopping”, and the most affected age categories are those between 17 and 34. In Romania, Telegram became the main platform associated with investment SCAMs, with 39% of cases.

Revolut warns that, despite the new technologies and security mechanisms, financial institutions cannot combat the growing volume of online fraud. Most fraud attempts start from social networks, not inside banking systems.

Collaborations between banks, DNSC and authorities

However, at present, there is already an active collaboration between the state institutions and the banking sector, through a program called “Online Security”, carried out in partnership with the Romanian Police, the National Cyber Security Directorate and the Romanian Banks Association. The program has both an educational and an operational component, meant to help banks develop internal reaction mechanisms in the face of digital fraud.

However, he recently explains for Adevărul, Dan Cîmpean, the director of the National Cyber Security Directorate, many of the fraudulent transactions are initiated by users, following the manipulation by social engineering techniques. The victims come to carry out transfers, sometimes hundreds or thousands of euros, without realizing that they are deceived, convinced of the veracity of the message received or the false identity of the interlocutor.

Although banks implement algorithms and complex systems that frequently manage to block such transactions, criminal methods are constantly changing. New mechanisms appear, difficult to anticipate, and sometimes the defense technology does not keep up. For this reason, it is essential that the victims immediately report the incident, either the bank or the police, so that the risks are known, investigated and, as far as possible, prevented. As long as only the victim is aware of what happened, the authorities cannot act effectively.

Globally, only 4% of victims recover their lost money

According to official DNSC data, Romania registers over 1,500 confirmed cyber incidents on the Moon. But this is just the visible part of the phenomenon. Experts estimate that only 15–18% of real attacks are detected, the rest passing unobserved or being ignored by the victims. Only in 2024, according to the most recent report published by the National Authority for Administration and Regulation in Communications (ANCOM), the number of security incidents that affected mobile and mobile internet connections in 2024 increased by 73.2% compared to the previous year, from 497 to 861.

In the age of AI, not the technology itself saves us, but how well we understand it. When everything can be forged, the safest investment remains prudence.