Audio Deepfake: Low-cost 3-second voice cloning, risks fraud and misinformation

A synthetic vocal track that faithfully replicates the nuances, intonation and timbre of a real person starting from three seconds of real voice recording and at low costs, even 30 euros. The evolution of generative artificial intelligence has made sophisticated systems possible which on the one hand offer accessibility tools for those who have lost the use of speech - the case of the actor Val Kilmer for the film Top Gun is emblematic - on the other they are redrawing the frontiers of cybercrime and scams such as the recent one against a Fideuram manager or involving celebrities whose image is manipulated. A topic that returned to the fore after Prime Minister Giorgia Meloni registered her voice at the European Union Intellectual Property Office to defend herself from AI fakes.

There is no precise figure for the number of cases of audio deepfake in the world, but sector data highlights a strong general trend. For security company McAfee, deepfakes (video, photo and audio) have increased by 900% in a year and the overall increase in fraud attempts has seen surges of more than 3,000% with attacks increasing by 700% year over year in 2026. More specifically, crimes related to voice cloning and audio deepfakes have increased by 1,300%.

Criminals can use a short snippet of a person's voice, sometimes just three to five seconds of audio is enough to get an 85% voice match between the original voice and the clone. In an analysis dated September 2025, Kaspersky also reported that it takes only 50 dollars to make a video deepfake and 30 for an audio one, about 400 times less than a few years ago. And several programs have been found on the dark web, often cheap, which allow these fakes to be created. According to a report by IdentifAI, the most common methods of deepfake attacks are video (45.6%), mixed media (25.2%), images (17.4%), and voice cloning (10.5%).

The manipulations have a political purpose in 24.6% of cases, 20.1% aim at fraud, 11.3% concern pornography, satire 3.2% and celebrities 9.3%. This is also why the creative world is organizing itself. The actor Matthew McConaughey has decided to patent his image to defend himself from the improper use of AI and several Hollywood stars such as Cate Blanchett and George Clooney have started a free public registry to give consent to the possible use of artificial intelligence for creative work, voice and image.

The United Nations International Telecommunication Union (ITU) has called on companies to use advanced tools to detect and eliminate disinformation and deepfake content to help counter the growing risks of election interference and financial fraud. An extremely urgent topic.

In fact, according to data reported by security.org, 70% of people said they were not sure they could distinguish between a real voice and a cloned voice, and that the most advanced automated detection systems suffer a 45-50% drop in accuracy when compared to real deepfakes compared to laboratory conditions.