It’s a change in proportions

According to FBI data, Americans lost more than $893 million last year to fraud involving artificial intelligence in one form or another: from voice cloning and automatically generated phishing to romance scams carried out by conversational agents capable of simulating empathy at industrial scale. However large the figure may seem, analysts consider it only a minimum threshold, because most victims never report the incident, either out of shame or because they believe there is no longer any way to recover their money. Aidapted has previously analyzed scams involving artificial intelligence, including the case of Simon, who lost $800,000 in a romance scam. What connects the cases below, however, is not the amount lost but the mechanism: in every situation, AI did not “break into” anything; it persuaded. And when the attack mechanism is trust rather than a technical vulnerability, the potential victim is no longer merely someone who was careless, but anyone. Let’s look at the three recent cases in detail.datelor FBISimon, cel care a pierdut 800.000 de dolari în urma unei escrocherii romantice

The Tutorial That Teaches You to Steal From Yourself

On September 14, blockchain investigations firm TRM Labs published a report describing one of the most ingenious forms of crypto fraud identified to date. Nine seemingly independent tutorials, published on YouTube between February and August, promised to teach users how to build, using an AI assistant, an automated crypto-arbitrage program supposedly designed to generate profits from price differences between trading platforms. The videos, produced with a visual consistency suggesting a single operation disguised under different creator identities, guided viewers step by step: open a wallet, copy the code, deploy the smart contract, and fund it.un raport

The problem lies precisely in the final step. What the victim actually installed was not a trading program, but a disguised trap: when the user pressed the start or withdrawal button, following exactly the steps indicated in the tutorial, the money was automatically sent directly to accounts controlled by the fraud operators. And this is what makes the case technically remarkable: the victim was not deceived through a fraudulent link, a fake webpage, or a suspicious wallet-permission request (the classic mechanisms detected by most protection systems), but implemented and funded the fraud with their own hands, convinced they were following legitimate technical instructions. TRM Labs identified 224 victims and 234 funded contracts, with a total of 274.6 ETH stolen (Ether, the native currency of the Ethereum blockchain), worth approximately $517,000, directed to six common addresses — a sign of a coordinated operation rather than isolated incidents.

The Face of a Financial Expert, the Lie You Cannot See

At the beginning of September, police in Northern Ireland confirmed that a single victim had lost £250,000 in a fraud that officials described as “sophisticated.” It all began with an AI-generated advertising video in which a well-known figure from the financial sector appeared, seemingly authentically, promoting an investment opportunity offering unusually high returns. The victim started with a modest sum, small enough to seem like a reasonable test of an opportunity rather than a risky bet. What followed, however, was a gradual manipulation strategy built patiently over time: conversations moved to WhatsApp, repeated pressure for increasingly larger sums, the creation of multiple online accounts, and finally remote access to the victim’s own computer, under the pretext of providing “technical assistance” with managing the investment.poliția din Irlanda de Nord a confirmat

It is essential to emphasize a detail that police explicitly highlighted: the AI-generated video was not the mechanism of the fraud, but merely the entry point, the gateway through which the credibility of a real person was lent to a completely false story. The rest of the scam relied on classic psychological manipulation techniques, applied with almost methodical consistency. British authorities have also confirmed that the likeness of public figures particularly associated with financial trust, such as Martin Lewis, the well-known British financial commentator, and high-profile entrepreneurs such as Steven Bartlett, has repeatedly been used without consent in such fake advertising campaigns.Autoritățile britanice au confirmat

The Voice at the Other End of the Line

On September 15, CNN reported the case of a Michigan couple, Brian and Wendy VanDoeselaar, who were preparing to close on a home purchase after more than three decades without such a transaction. In the hectic days before signing the paperwork, when payment instructions circulate quickly and time pressure is at its highest, the couple received an email and, later, a phone call that appeared to come from the loan officer they had been working with for weeks. The displayed phone number matched a known contact, and the voice on the other end sounded exactly like the real person. Brian followed the instructions and transferred $66,000 — the exact, non-rounded amount of funds required to close the transaction, a detail that paradoxically made the scam more credible precisely because it did not look like an invented figure.CNN a relatat

Only days later, after receiving no confirmation that the money had arrived, did the couple notice an apparently minor error in an email address: “UnitedsMortgages” instead of the correct name, “UnitedMortgage” — a single “s” inserted with a precision that says a great deal about the attackers’ attention to detail. The transferred funds went to an account that was closed shortly after the fraud, consistent with a well-documented pattern: stolen funds are typically redirected immediately to crypto or overseas bank accounts, and experts estimate that after 48 hours, the chances of recovery become almost negligible.

What Has Changed

The three cases, although apparently different in form, collectively describe a single fundamental transformation in the world of fraud: visual and audio evidence — the image of a familiar face, the timbre of a familiar voice, the coherence of a well-made technical tutorial — have ceased, on their own, to be reliable grounds for trust. For decades, the human instinct naturally associated “I see it, I hear it, therefore it is real” with a reasonable level of safety. Generative AI has eroded precisely that connection, and the speed with which the technology is becoming more accessible and more convincing suggests that the phenomenon is only beginning to spread.

The lesson security specialists repeat, in different forms, in each of these cases, is surprisingly simple in theory and difficult to apply in practice, precisely because it runs against our natural reflex toward urgency: any unusual financial request, no matter how convincing its source may seem, deserves to be verified through a channel completely independent of the one through which it was received. Do not call back the number provided in the message; use one you already know. Do not continue a conversation on a platform suggested by the person contacting you; reach out to them through means you control yourself. And when someone insists on speed, secrecy, or the impossibility of verifying the information through another channel, that pattern of pressure matters more than any other apparent proof of authenticity because, as these three stories show, apparent authenticity has become, in 2026, the cheapest resource to manufacture for anyone willing to lie.

Sources: