Cloning a voice no longer takes hours of recording: why older adults are the easy target for fraudulent AI

By Telediario México · July 1, 2026.

Ramón Morales, vice president of Emerging Technologies at CANIETI, has put qualitative figures to something that at Zendoric we have been flagging for months as the most urgent AI risk of the present: not distant superintelligence, but the industrialization of everyday fraud. His warning is concrete and technical: cloning a person's voice or image no longer requires the hours of recordings and photographs that models from barely a couple of years ago demanded. All it takes is publicly available apps and a few minutes to generate fake content —audio, image or video— capable of deceiving a victim or sustaining an extortion attempt. According to information Morales attributes to the director of Jalisco's C5, the number of frauds linked to these tools has shown a significant increase in the region, and the sector identifies older adults as the most exposed group, precisely because of their lesser familiarity with the current reach of the technology.

The fact itself is modest in origin —an institutional statement to a regional Mexican outlet— but it fits a trend documented in parallel in much larger markets: specialized coverage of financial fraud (such as that followed by outlets like Accounting Today in the United States) has spent months describing the same phenomenon from the banking and accounting angle. It is no coincidence that it appears simultaneously in Jalisco and in the Anglo-Saxon financial press: the barrier to entry for producing a convincing voice or image deepfake has collapsed everywhere at once, because the tools that make it possible are the same commercial apps, not handcrafted software from criminal labs.

As sector context, projections on AI-powered banking fraud speak of a jump from some $23 billion in 2025 to more than $58 billion by 2030, an increase of over 150% in five years. That global figure explains why institutions like CANIETI choose to speak now, rather than wait for the problem to mature: the cost of not warning in time grows non-linearly, and older adults —with less prior digital exposure, greater trust in the recognized voice of a relative and fewer cross-verification networks— are statistically the cheapest link for a scammer to break.

Our reading is that this kind of news, however local or anecdotal it may seem, is the best available thermometer of how generative AI moves from the lab to the street before there is social infrastructure to absorb it. The gap is not technological —deepfake detectors and dual-channel verification protocols already exist— but educational and generational: the same ease that lets a grandchild send a funny audio clip to their grandmother lets a scammer simulate that grandchild's voice urgently asking for money. Morales's recommendation to involve families and educational institutions is not a whim, it is the only genuinely available short-term lever, because regulation and technical detection will always be one step behind the app of the moment.

In the long run, however, it is worth keeping perspective: the same AI that today clones voices for criminal purposes is, in essence, the technology that within a few years will help diagnose diseases before they show symptoms or accompany those same older adults with assistants capable of detecting scam patterns in real time, alerting before the call ends. The mismatch we are experiencing now —offensive capability running faster than everyday defense— is exactly the kind of transitional friction that was to be expected: painful, real, and disproportionately aimed at the most vulnerable, but not representative of where the technology is heading if it is governed sensibly. The immediate challenge is not to slow down AI, but to educate the population that knows it least before the cost of the lag is paid in consummated frauds.