The number of attempts to defraud people and institutions using synthetic audio, video, and documents has climbed steeply over the past few years, and the pace of that increase has become one of the most cited warning signs in the fraud-prevention field. Where a deepfake once meant a novelty clip of a celebrity, the term now describes a working criminal tool: a fabricated voice on a phone call, a manipulated video on a conference screen, or a forged identity document used to open an account. The scale of the jump is what has regulators, banks, and security teams paying attention.
The 2,137% figure and what it counts
The most frequently repeated statistic holds that detected deepfake fraud attempts have risen more than 2,000% over a three-year span, a compilation that puts the precise increase at 2,137%, according to an analysis of AI-driven fraud statistics. The same body of data suggests that roughly one in fifteen detected fraud cases now involves some form of synthetic media, a share that would have been negligible only a few years earlier. Those figures measure attempts and detections rather than successful thefts, but the trajectory is the point: the technique has moved from the fringe of cybercrime toward its mainstream.
It is worth reading such percentages carefully. A jump measured off a very small starting base can look enormous in relative terms, and detection counts partly reflect better tools for spotting fakes. Even allowing for that, analysts across the sector agree the underlying volume is rising quickly, and the direction is not in dispute.
Why synthetic media got cheap and abundant
The growth tracks the falling cost of the tools that produce fakes. Voice synthesis that once required expensive software and clean recordings can now be run from a subscription website using a short clip, and comparable advances in video and image generation have made convincing forgeries far easier to create, as documented in reporting on the spread of AI cloning tools. The same accessibility that lets a small business make a marketing video lets a scammer fabricate a chief executive issuing a payment order.
Criminals have adapted their business models accordingly. Fraud rings can automate large volumes of attempts, target both consumers and corporate finance departments, and stitch synthetic voices or faces into schemes that previously relied on stolen passwords alone. The result is a broader attack surface: not just impersonation calls to families, but forged onboarding documents, manipulated video in remote identity checks, and fabricated recordings used to pressure employees.
Where the fakes show up in real money transfers
The costliest incidents tend to hit organizations rather than individuals. Finance staff have been deceived by video calls in which colleagues appeared to authorize transfers, and identity-verification systems that rely on a selfie or a short video have been probed with synthetic faces. On the consumer side, the same technology powers the emergency-money calls in which a relative’s cloned voice begs for help, a scheme that succeeds by triggering panic before the target can verify anything.
Across both, the common failure is trusting a single channel. A voice, a face, or a document that looks authentic is no longer sufficient proof of who is on the other end, because each of those signals can now be manufactured on demand.
How institutions and families are trying to verify identity
The defenses that hold up best do not depend on catching a flaw in the fake. Financial institutions are layering additional checks, callbacks to known numbers, multi-person approval for large transfers, and liveness tests designed to be hard for a synthetic image to pass, so that no single spoofable signal can move money on its own. Households are being urged toward the same principle in miniature. Consumer-protection officials recommend agreeing on a private code word that any genuine caller must supply before an urgent money request is treated as real, guidance laid out in the Federal Trade Commission’s alert on harmful voice cloning. A fabricated voice cannot produce a secret it never received.
The shared lesson is verification through a second, independent channel. Calling a person back on a trusted number, confirming a payment request in person or through a separate system, and refusing to be rushed all defeat a fake that would otherwise pass at a glance.
What a rising curve does and does not mean
A headline percentage is a snapshot of momentum, not a forecast, and the eventual ceiling depends on how fast defenses improve alongside the tools. Detection software, watermarking of AI-generated content, and stricter verification standards are all advancing, and each raises the cost of a successful fake. What the multi-year climb makes clear is that synthetic media has become a durable part of the fraud landscape rather than a passing novelty, and that the habits built to counter it, verify first, trust no single signal, will matter for years regardless of where the next figure lands.
This article was researched and written with the assistance of AI and reviewed by an editor prior to publication.
More from Morning Overview
