The FBI's Internet Crime Complaint Center recorded 22,364 complaints referencing artificial intelligence in 2025, totaling nearly 900 million dollars in reported losses, and researchers tracking crypto specific impersonation scams have observed growth exceeding 1,400 percent year over year. Nothing about the underlying scam patterns covered elsewhere on this site, fake investment platforms, romance scams, phishing, has fundamentally changed. What has changed is how convincing, fast, and cheap it now is for a scammer to produce the material that makes those old patterns work.
Why AI lowers the barrier to running a convincing scam
Running a scam has always required some combination of a believable script, a believable voice or face, and enough volume to make the numbers work despite a low success rate. Generative tools now automate large parts of all three. A single operator can produce fluent, personalized messages in multiple languages instantly, generate a synthetic voice from a few seconds of audio, and create video of a face that never said the words being spoken. None of this requires technical skill anymore. Consumer tools that were experimental a few years ago are now accessible enough that a small team can run operations that previously required a much larger staff.
Voice cloning and the new vishing threat
Voice cloning tools can now produce a convincing replica of someone's voice from as little as three seconds of audio, reaching roughly 85 percent accuracy according to researchers tracking the technology. This has turned old fashioned phone based fraud into something far more dangerous. A call that sounds exactly like a family member in distress, a colleague requesting an urgent transfer, or an executive authorizing a payment carries a level of immediate credibility that a text message never could. Surveys have found that roughly one in ten people report having received a message from a cloned voice, and a majority of those who did lost money as a result. Enterprises hit by voice based fraud attacks report average losses in the hundreds of thousands of dollars per incident.
Deepfake video and celebrity endorsement scams
Fabricated video has become common enough in crypto fraud that it now has its own recognizable patterns, most visibly in fake celebrity endorsements for investment platforms or token launches. These videos frequently reuse real footage of a well known figure with the audio replaced, or generate an entirely synthetic likeness delivering a scripted pitch. Our guides on Elon Musk impersonation scams, celebrity impersonation crypto scams, and deepfake crypto scams cover these specific patterns in detail, since the underlying technique of borrowing a trusted public face to sell a fraudulent opportunity now shows up across nearly every major platform.
AI generated romance and chat personas at scale
Perhaps the most consequential shift is happening quietly, inside the long running conversations at the center of pig butchering and romance scams. Where a single scammer once managed a handful of ongoing relationships manually, generative chat tools now allow one operator to maintain dozens of consistent, emotionally responsive conversations simultaneously, each one remembering personal details and adjusting tone appropriately. Our guide on how pig butchering scams work explains the underlying scheme these conversations lead toward, which has not changed. What has changed is the volume of victims a single operation can now realistically manage at once.
Fake recruiters and AI written job offers
The same tools are showing up in recruiting related scams, producing fluent, professional sounding job offers, interview scripts, and even synthetic video interviews that make a fabricated employer feel legitimate. This overlaps closely with the task and job scam patterns already common on messaging platforms, now delivered with a level of polish that used to require an actual company behind it. A candidate might now go through what feels like a full interview process, including a live video call with a synthetic interviewer, before ever being asked to pay for training materials or deposit cryptocurrency to unlock paid work, a pattern covered in more detail in our guide on WhatsApp job and task scams.
AI generated fake support chat
Fraudulent customer support has also become more convincing. Chatbots trained to mimic the tone and structure of a real exchange or wallet support team can now sustain a fluent, patient sounding conversation that used to require a human operator, guiding a target step by step through installing remote access software or revealing a seed phrase under the pretense of resolving an account issue. Because the responses feel personalized and the conversation can continue indefinitely without the delays a human scammer might introduce, targets sometimes stay engaged with a fake support conversation far longer than they would with an obvious script.
A realistic walkthrough of an AI enabled voice scam
A typical case begins with the attacker harvesting a short audio sample of the intended impersonation target, sometimes from a public video, a voicemail greeting, or a social media clip. That sample is fed into a voice cloning tool, and within minutes the attacker has a synthetic voice capable of saying anything typed into the software. The attacker then places a call to a family member, a colleague, or a member of a company's finance team, using the cloned voice to describe an urgent situation, an arrest requiring bail, a medical emergency, an executive needing an immediate confidential wire, that discourages the person on the receiving end from taking time to verify. Because the voice itself sounds correct, the emotional response often overrides the instinct to question the story, and the request, frequently for cryptocurrency specifically because it settles quickly, is fulfilled before anyone thinks to call the actual person back on a known number.
Recognizing when a loved one may be targeted
Because AI enabled scams often move faster than older scam patterns, warning signs can be easy to miss until money has already moved. A family member who receives an unusual, urgent call asking for money or a cryptocurrency transfer, especially one that discourages hanging up to verify or insists on secrecy, should always be encouraged to pause and call back on a number they already have saved, regardless of how convincing the voice sounded. If an older relative mentions a distressing call from a grandchild or family member in trouble, checking in directly with that family member through a separate channel before any money moves can prevent a loss that would otherwise be irreversible.
AI has not created a fundamentally new category of crypto scam. It has removed the cost and skill barriers that used to limit how many people a single scammer could convincingly target at once, which is the actual driver behind the sharp rise in losses.
The numbers so far
Blockchain analytics firms estimate that at least fourteen billion dollars reached identified crypto scam addresses in 2025, with that figure expected to climb as more illicit addresses are identified retroactively. Crypto specific impersonation scams, many involving fabricated video or audio of trusted figures, grew by more than 1,400 percent over the same period. These are still early numbers relative to how new the underlying tools are, and most researchers expect the trend to continue rather than plateau in the near term.
Practical signs that content may be AI generated
Detection is becoming harder as the underlying tools improve, but a few practical checks still catch a meaningful share of fabricated audio and video. Unnatural blinking patterns, lighting that does not quite match between a person's face and their surroundings, and audio that sounds slightly flat or oddly paced compared to how the person actually speaks in verified footage are all still common in current generation deepfakes. For voice calls specifically, background noise that never changes, a slight delay or robotic quality under stress, or a caller who avoids answering an unexpected, specific personal question can all be signs worth pausing over, though none of these signs should be relied on alone given how quickly the technology continues to improve.
How to adjust your own verification habits
- Treat an urgent voice or video call requesting money or a crypto transfer as unverified until confirmed independently, even if the voice or face sounds and looks exactly right
- Establish a verification phrase with close family members or business partners that would not appear in any public recording of their voice
- Call back using a number you already have saved, never one provided during the suspicious call itself
- Be skeptical of any celebrity endorsed investment opportunity, since public figures rarely promote specific crypto platforms directly, and video alone is no longer proof of authenticity
- Slow down deliberately whenever a message or call creates urgency, since urgency is the one ingredient nearly every version of this fraud still depends on
- Limit how much personal audio and video of yourself and close family members is publicly available online, since source material is what makes cloning possible in the first place
- Ask an unexpected question during a suspicious call that only the real person could answer naturally, rather than something that could be guessed or found online
Frequently asked questions
Researchers have found that as little as three seconds of audio can produce a voice clone with roughly 85 percent accuracy using current consumer tools. This means any public video, voicemail greeting, or social media clip can potentially be used as source material.
Treat any celebrity endorsement of a specific crypto platform or token as false by default, since well known public figures rarely promote individual investment products this way. Video quality alone is not a reliable way to detect a fake, since modern deepfakes can look and sound convincing even under close attention.
Both. Voice cloning has been used to impersonate executives authorizing urgent wire or crypto transfers, and businesses report average losses in the hundreds of thousands of dollars per incident when this succeeds. Our guide on {{link:business-email-compromise-crypto|business email compromise and cryptocurrency payment fraud}} covers the related pattern of impersonation used against companies specifically.
It has made them more scalable rather than fundamentally different. A single operator can now manage many more simultaneous, personalized conversations using generative chat tools, which increases the total number of people a scam operation can realistically target at once.
Detection tools exist and are improving, but none are reliable enough yet to depend on during a live, high pressure call. The more dependable approach is procedural rather than technical, verifying any urgent financial request through a separate, previously established channel rather than trusting the call itself, regardless of how convincing the voice or face sounds.
Sources and further reading
- Cryptocurrency and AI Scams Bilk Americans of Billions · Federal Bureau of Investigation
- AI Voice Cloning Fraud Statistics 2026 · SQ Magazine