Can you still trust a voice on the phone?

The claim travels as a pair of statistics attached to one conclusion. AI voice cloning, it says, has become the dominant fraud vector. Deepfake complaints to the Federal Trade Commission are up 1,200 percent, and 40 percent of business email compromise attacks now use AI voice or video. No author, date or document travels with it. It does name a federal agency, which makes it sound like something a reader could check.⁴,¹

Neither statistic appears in the FTC and FBI records checked for this piece, and the claim attributes only the first to an agency. Voice cloning is a real and documented fraud tool. The FBI's 2025 ledger does not show it as the dominant one. Complaints that referenced AI accounted for about 4 percent of reported losses.¹

Does the FTC count deepfake complaints?

The FTC's annual accounting of fraud reports is the Consumer Sentinel Network Data Book. The 2024 edition was published in March 2025. It states that it is based on unverified reports filed by consumers and is not a consumer survey.⁴ In the portion that was read, its 29 report categories include no deepfake or AI voice category.⁴ The appendices could not be read in full, so that absence holds for the main tables only.

On 15 June 2026, the FTC announced that people reported losing $3.5 billion to imposter scams in 2025. Nearly one in three fraud reports concerned imposter scams.³ The announcement says nothing about deepfakes, voice cloning or a 1,200 percent rise.³

The FTC's prepared statement to the Joint Economic Committee on 25 March 2026 gave the 2025 totals: 3 million fraud reports and $15.9 billion in reported losses.⁵ It does not break out deepfake or voice-cloning complaints. It mentions the agency's 2023 to 2024 Voice Cloning Challenge only in passing, in a paragraph on industry collaboration, with no complaint data.⁵ A 1,200 percent increase needs a published series to measure against. None of these three FTC documents contains one in the portions read.

What does the FBI count?

The FBI's Internet Crime Complaint Center published its 2025 report in 2026. It logged 1,008,597 complaints and $20.877 billion in reported losses.¹ Complaints tagged as AI related numbered 22,364, with losses of $893,346,472.¹ That is about 2.2 percent of complaints and about 4.3 percent of dollars.

The largest loss categories were investment fraud, at $8.65 billion, and business email compromise, at about $3.05 billion.¹ AI-related losses across every category combined were roughly a tenth of investment fraud alone. The ledger does not show AI, let alone voice cloning specifically, as the dominant vector.

The Senate Special Committee on Aging repeated the same figures in its July 2026 report. It cited nearly $21 billion in 2025 cybercrime losses, with $893 million tied to AI-enabled scams across more than 22,000 complaints.² It drew its fraud data from the IC3 and the FTC. It added no independent count of deepfake complaints.²

Where does the 40 percent come from?

The IC3 report addresses AI within business email compromise directly. It says voice cloning can be used to request wire payments. It also says there are multiple BEC tactics and not all are AI-enabled. In 2025, businesses reported losses over $30 million to BEC scams involving AI.¹ Against roughly $3.05 billion in BEC losses, that is at least about 1 percent by value. That measures dollars, not the share of attacks.¹

No primary record located for this piece says 40 percent of BEC attacks use AI voice or video. The nearest number in circulation is a security vendor's blog post (Hoxhunt) relaying a VIPRE Security Group estimate, via SC Media, that about 40 percent of BEC emails in Q2 2024 were AI-generated, based on a sample run through AI-text detectors.¹¹ That measures how a message was worded, using AI-text detectors on a sample of unstated size. It does not measure cloned voices or faked video, and the underlying VIPRE report was not reviewed for this piece. The same post says AI voice clones and video deepfakes are "still relatively rare but growing."¹¹

Is the FBI figure the real share?

No. The AI tag counts only what complainants reported, and the IC3 report does not present it as the full share.¹ The report describes AI-enabled content as increasingly hard to detect.¹ The true share may be higher than the tagged figure, because victims who never suspected a synthetic voice would not have flagged AI. No record measures by how much, and nothing in the record supports 40 percent.

Is voice-clone fraud documented at all?

Yes. The Federal Communications Commission adopted a ruling on 2 February 2024 (FCC 24-17, released 8 February). It held that AI technologies such as voice cloning fall within the Telephone Consumer Protection Act's existing restriction on artificial or prerecorded voice messages, which requires the called party's prior express consent.⁷ The ruling is a legal interpretation. It does not measure how common the fraud is, and it does not stop callers who ignore the law.⁷

On 13 November 2024, the Treasury's Financial Crimes Enforcement Network issued alert FIN-2024-Alert004. It reported that criminals have used generative AI to impersonate executives or trusted employees and instruct victims to send large payments to accounts the scammers control.⁶ The alert focuses mainly on deepfaked identity documents used to open or access accounts. It gives no volume or percentage figures.⁶

A 2025 IC3 brochure produced with the ABA Foundation says imposter scams are rising and that criminals use deepfakes to gain trust.¹⁰ Its headline figures, more than 4.2 million reports and $50.5 billion since 2020, cover all fraud reported to the FBI, not deepfake fraud.¹⁰ The World Economic Forum's Global Cybersecurity Outlook 2026 says criminals use generative AI to produce deepfake audio and video that can evade detection systems and human scrutiny.⁹ That report rests on surveys and focus groups. It reflects perceptions, not incident counts.⁹

Can a listener tell the difference?

Not reliably. In a study published in PLOS ONE on 2 August 2023, researchers at UCL played speech clips to 529 online participants. Listeners correctly identified the deepfakes 73 percent of the time, with no difference between English and Mandarin.⁸ The test used short clips from one speaker per language and publicly available text-to-speech models. It did not test live calls, familiar voices or pressure to act quickly.⁸ It shows that the ear is a weak detector. It says nothing about how often fraud uses cloned voices.

The Senate Aging report says the FBI recommends a family code word. Its endnotes list an FBI IC3 public service announcement from December 2024 that gives that advice, though the report does not tie the note to the sentence explicitly.²