Research Report

AI Voice Cloning Fraud: The Real Numbers, Sourced

"AI voice scams cost Americans $3.5 billion" is one of the most-repeated statistics in cybersecurity coverage right now. It's also not quite what it sounds like. Here's what the primary sources -- Deloitte, Pindrop, the FBI and the FTC -- actually say about the scale of AI voice cloning fraud, including which widely-cited numbers don't hold up to a source check.

Botnira Research · Reviewed by Kamaljeet Singh Sidhu, Founder & CEO of Botnira · Published September 6, 2026 · 8 min read

Key findings

What's actually verified vs. what's commonly misquoted

📈

+1,300% in 2024

Rise in deepfake fraud attempts against enterprises, per Pindrop's 2025 Voice Intelligence & Security Report -- from ~1/month to ~7/day.

🎙️

3 seconds

Audio needed to produce an 85%-match voice clone, per McAfee's "Artificial Imposters" study of 7,000 people in 9 countries.

💵

$25M (real case)

Wired by a Hong Kong Arup employee after a deepfake video call impersonating the company's CFO and colleagues, January 2024.

🚩

$3.5B is NOT AI-specific

The FTC's 2025 figure covers imposter scams broadly. The FTC's own release does not isolate AI or voice cloning as a category.

What the primary sources actually report on scale

Pindrop's 2025 Voice Intelligence & Security Report, published June 12, 2025, is the most detailed enterprise-facing data point available: deepfake fraud attempts against enterprises rose 1,300% in 2024 -- from roughly one attempt a month to about seven a day. The same report found synthetic-voice attacks specifically up 475% in insurance, 149% in banking and 107% in retail, with contact-center fraud attempts occurring, on average, every 46 seconds. Pindrop projects $44.5 billion in contact-center fraud exposure for 2025 -- a projection from a vendor that sells fraud-detection tools, worth reading with that context, though the underlying attempt-volume data is the most granular published on this specific threat.

Deloitte's Center for Financial Services, in a May 29, 2024 analysis, projects generative-AI-enabled fraud losses in the US could reach $40 billion by 2027, up from $12.3 billion in 2023 -- but this figure covers generative AI fraud broadly (deepfake documents, synthetic identities, phishing content), not voice cloning in isolation. The FBI's Internet Crime Complaint Center (IC3) 2025 Annual Report recorded $893 million in AI-attributed fraud losses across 22,364 complaints, out of $20.877 billion in total cybercrime losses that year -- and explicitly notes voice cloning is "now being layered" into business email compromise and grandparent/distress scams, without breaking out a voice-specific dollar figure.

Sources: Pindrop, 2025 Voice Intelligence & Security Report; Deloitte Center for Financial Services, May 29, 2024; FBI IC3 2025 Annual Report (PDF).

Why the "$3.5 billion AI voice scam" figure needs a caveat

The FTC's June 2026 press release, covering 2025 data, reported $3.5 billion in consumer losses to imposter scams -- over 1 million reports, up 19% year over year. This is a real, sourced FTC figure. What it is not is an AI-voice-specific statistic: the FTC's own release does not isolate AI-generated or voice-cloned calls as a distinct sub-category within that total. A number of secondary outlets have nonetheless repeated this figure framed as "AI voice scam losses," which is a conflation worth catching before repeating it further -- the underlying imposter-scam category includes calls, emails, texts, and in-person schemes that have nothing to do with AI voice cloning at all.

The same caution applies to occasionally-seen claims that treat Deloitte's $40 billion generative-AI-fraud projection as a voice-cloning-specific number, or that cite the FBI's $893 million AI-fraud figure as isolated to voice cloning. None of these primary sources break their totals down by voice cloning specifically -- if a piece of content asserts otherwise, it's worth checking whether it traces back to an actual primary source or just to another blog repeating the same unsourced framing.

Real, named incidents -- and where the record is thinner than you'd expect

Two well-documented, named cases anchor most coverage of this topic. In Hong Kong, January 2024, an employee at the engineering firm Arup wired $25 million after a video conference call where deepfaked video and voice impersonated the company's CFO and several colleagues -- technically a deepfake video call rather than pure voice cloning, but frequently cited in voice-fraud coverage. Separately, in the US, Jennifer DeStefano testified before the Senate Judiciary Committee on June 13, 2023 that scammers used an AI-cloned version of her daughter's voice in a fake kidnapping call demanding ransom -- a genuine, sourced consumer-fraud incident.

What's notably harder to find: a verified, named case of voice-cloning fraud executed specifically against a bank's or call center's voice-biometric authentication system, with a confirmed dollar loss attached. Vendors like Pindrop discuss this exposure in aggregate statistical terms (attempt volumes, percentage increases), but a concrete, publicly reported "bank lost $X to a cloned-voice authentication bypass" case was not found in this research pass -- worth treating the risk as real and rapidly growing based on attempt data, while being precise that headline-grabbing named incidents to date have mostly been video-deepfake or family-impersonation scams rather than institutional voice-biometric breaches.

Sources: Trend Micro, Arup deepfake case; CNN, Feb 4, 2024; Jennifer DeStefano, Senate testimony (PDF); McAfee, "Artificial Imposters".

FAQ

Frequently asked questions

How much audio does it take to clone someone's voice?

McAfee's "Artificial Imposters" study found just 3 seconds of audio produced a voice clone with an 85% match to the original voice, rising to 95% with more training audio -- based on a survey of 7,000 people across 9 countries.

Is the widely-cited $3.5 billion AI voice scam loss figure accurate?

Not as usually reported. The FTC's actual 2025 figure is $3.5 billion in imposter-scam losses broadly -- the FTC's own release does not isolate AI or voice cloning as a distinct sub-category. Secondary sources have loosely relabeled this as an "AI voice scam" statistic, but that framing doesn't trace back to the FTC's own report.

How much did deepfake fraud attempts against enterprises actually increase?

Pindrop's 2025 Voice Intelligence & Security Report found deepfake fraud attempts against enterprises rose 1,300% in 2024, from roughly one per month to about seven per day, with synthetic-voice attacks up 475% in insurance and 149% in banking specifically.

Kamaljeet Singh Sidhu

Founder & CEO of Botnira and CEO of The DigiSparrow. Reviews the independent industry research published on this hub.

Read full bio →