AI fraud surged 1210% just last year.¹

Listen now

[!AUDIO]https://cdn.prod.website-files.com/6a809521fc6b7dfa71f2e253/6aa17bfe094e7b41e758c5c6_AI-Fraud-Guide_Chapter-2.mp3[/]

2025 was the year of AI attacks.

According to Pindrop internal data, AI fraud (or non-live fraud) surged 1210% by December 2025.¹ From this, it’s clear that attackers are rebuilding their operations around AI. But why? Because it’s cheaper, faster, harder to detect, and startlingly scalable.

With automated models, today’s attackers don’t get tired, don’t act on emotion, and don’t reuse the same face or voice twice. Attackers can train models with rigor, and once trained, these models work non-stop to exploit your vulnerabilities.

‍

“We’ve seen attacks in the private wealth market, call centers, or even IT help desks. It’s everything from attacking the clients and customers directly to the people interacting with those customers.” Matthew Miller, Principal Partner, KPMG

What does an AI scam look like?

In voice channels, it often starts quietly. Bots hit the Interactive Voice Response (IVR) system, not to drain funds immediately, but to learn. They identify which prompts trigger security checks and attempt to validate breached data. Their goal first and foremost is reconnaissance. Later, those same bots—now smarter—come back armed with knowledge about weak points, setting up a much more effective fraud attempt.

Listen to these calls. Can you tell the difference?

This is a real person.

[!AUDIO]https://cdn.prod.website-files.com/6a809521fc6b7dfa71f2e253/6aa80f7c2195275a4a542454_Sam-Original_AI-Fraud-Spike-Guide.mp3[/]

This is an AI clone.

[!AUDIO]https://cdn.prod.website-files.com/6a809521fc6b7dfa71f2e253/6aa17c5399ac482b531d057c_Sam-ElevenLabs_AI-Fraud-Spike-Guide.mp3[/]

In other real-time channels like video meetings, the scam is often direct and bold. An employee gets pulled into a last-minute meeting. On screen is a convincing impersonation of their CFO. The CFO’s face, voice, and mannerisms are realistic—the employee sees no cause for alarm. The request is urgent: a sensitive transaction, a delayed payment, a problem that cannot wait. The fake CFO uses the same old social engineering tactics, now backed with video and audio credibility. The attacker applies just enough pressure to bypass normal checks: “I’ll explain later” and “I need you to take care of this right now.” By the time the employee realizes the meeting was deceptive, the money is gone.

Humans catch AI only ~50% of the time.²

“Human ears and human eyes are just not enough. They’re rendered ineffective at determining what’s real, who’s real, and who isn’t.” Amit Gupta, VP, Product Management, Pindrop

Even with awareness, humans remain the weakest link, especially when urgency or perceived authority enter the conversation. Understanding that weakness, attackers now generate highly realistic human videos and voices, deliberately making small talk and sounding patient, polite, and real. That perceived real-ness builds exploitable trust.

In a recent academic study, a synthetic voice bot called ViKing successfully extracted sensitive information from 52% of participants using AI-generated speech.³ Even more concerning: when participants were explicitly warned that synthetic bots were common, they still shared information 33% of the time.³ Awareness helps but the data is clear: training alone isn’t enough to stop AI-assisted social engineering.