According to the FBI’s Internet Crime Complaint Center (IC3) 2025 annual report, for the first time in 26 years, “AI-related scams” have been listed as a separate crime category, with more than 22,000 complaints nationwide. Consumer Reports tested six AI voice-cloning tools, including Descript, and found that four of them allow users to generate cloned audio by simply checking a box that says “I have the right to clone this voice segment,” without any verification.
FBI IC3 2025 Annual Report: AI-Related Scams First Time Listed as a Crime Category
According to the FBI IC3 2025 annual report published in April 2026, the following are key complaint and loss figures:
AI-related scam complaint count: more than 22,000 cases, the first time the FBI has listed them as a standalone crime category in 26 years
Adjusted losses: over $893 million
Losses of victims age 60 and over: $352 million (about 39% of total AI scam losses)
Year-over-year growth rate of cybercrime losses across the U.S.: 26%; losses for the 60-and-over group totaled $7.7 billion for the year (up nearly 60% year over year)
FTC estimated true annual losses: possibly as high as $81.5 billion (including the “dark figure” of unreported cases)
Interpol 2025 global financial scam losses: $442 billion, with AI scams making 4.5 times more than traditional methods
In the report, the FBI acknowledges that the complaint figures are only a “floor price,” as most victims do not know that the person they spoke with was an algorithm.
Consumer Reports 2025 Review of Six Voice-Cloning Tools: Current Anti-Abuse Protections
After Consumer Reports reviewed six voice-cloning tools—Descript, ElevenLabs, Lovo, PlayHT, Resemble AI, and Speechify—in 2025, it found that most platforms lack meaningful anti-abuse protections. Of the six, four require only that users independently check a claimed authorization statement—“I have the right to clone this voice segment”—to generate the audio, with no verification process.
ElevenLabs is a relatively stricter example: it uses multi-layer defenses such as a classifier, traceable records, and the ability to block certain people’s voices. However, Consumer Reports said these mechanisms are almost entirely post-hoc.
On the technical side, a voicemail recording, a birthday video, or a short clip from social media can be used as training material for a voice-cloning model, with original recordings as short as 3 seconds. Hany Farid, a deepfake forensics expert at the University of California, Berkeley, said in a June 2026 interview with The New York Times that he can no longer pass his own synthetic speech recognition tests.
UK PSR October 2024 Mandatory Reimbursement Scheme
Beginning in October 2024, the UK Payment Systems Regulator (PSR) requires banks to fully reimburse victims of authorized push payment (APP) scams. Both the receiving and sending banks each bear half the responsibility, with a case cap of £85,000. After the scheme took effect, the reimbursement rate for scammed amounts rose from 65% in the past to 89%.
On legislative progress: related obligations under the EU AI Act began taking effect from 2025 to 2026; Tennessee’s ELVIS Act (Ensuring Likeness Voice and Image Security Act) requires written prior consent for any copying of an individual’s voice.
An executive at voice-safety company Pindrop, Amit Gupta, said in a public statement: “The goal of a scam is not to perfectly replicate a voice, but to create enough emotional uncertainty and urgency so that victims act before they verify.”
FAQ
When did the FBI first list AI scams as a separate crime category?
In the FBI IC3 2025 annual report published in April 2026, “AI-related scams” were listed separately as a standalone category for the first time in 26 years, with more than 22,000 complaints in a year and adjusted losses of over $893 million.
What anti-abuse issues do AI voice-cloning tools have right now?
According to Consumer Reports’ 2025 review, of the six major tools, four require users to generate cloned audio by simply checking an authorization statement themselves, with no substantive verification steps. Although ElevenLabs has multi-layer protections, Consumer Reports said most of these mechanisms are geared toward post-hoc investigation.
How effective is the UK PSR mandatory reimbursement scheme?
Starting in October 2024, the UK PSR mandated that banks fully reimburse APP scams, with a cap of £85,000; the receiving and sending sides each take half. After the scheme began, the reimbursement rate rose from 65% to 89%; the specific reimbursement conditions for individual cases depend on the PSR’s official rules and each bank’s policies.
