AI has become a tool for a new generation of fraudsters: how deepfakes are stealing millions of dollars - research

• TRM Labs recorded a 263% increase in losses from deepfake fraud in the early months of 2026 compared to the total for the entire previous year.
• In particular, criminals are cloning voices, faces, and documents.
• Among the recognition recommendations are certain visual, audio, and behavioral characteristics.

In 2026, deepfakes transformed from an experimental technology into a tool for fraud, theft of funds, bypassing KYC, and creating illegal content. According to TRM Labs, the AI-in-Crime Adoption Index rose from 28 points in 2024 to 54 in 2026, and losses from deepfake fraud in the early months of 2026 already exceeded the total for all of 2025 by 263%.

According to TRM's 2026 AI-in-Crime Adoption Index, deepfake-scam losses in early 2026 already exceeded all of 2025 by 263%. Consumer tools can now clone a voice from seconds of audio, making executive impersonation and "grandparent scams" easier to run at scale.

New on the TRM... pic.twitter.com/j0hj7Ft4B0
--- TRM Labs (@trmlabs) September 22, 2026

Amid the decreasing costs of generating and distributing accessible AI tools, the technology simultaneously creates risks for the financial sector, cryptocurrency platforms, businesses, and ordinary users.

How deepfakes are used in fraud

One of the most well-known cases occurred in January 2024 at the engineering company Arup. An employee was communicating during a video conference with five "colleagues," one of whom was the company's CFO from the UK. In reality, all the interlocutors, except for the employee himself, were deepfakes. Believing the instructions received, he approved 15 transfers totaling $25 million to accounts in Hong Kong.

According to Deloitte's estimates, losses from fraud using generative AI in the US could rise from $12.3 billion in 2023 to $40 billion by 2027. The FBI also repeatedly warned in 2025-2026 about the use of synthetic videos and voices to impersonate officials, company executives, and law enforcement officers.

Among the main schemes:

• impersonation of executives and relatives to obtain transfers, authentication codes, or confidential information;
• voice cloning in modern variations of "grandchild scams";
• romantic scams where AI generates photos, videos, and helps maintain multiple dialogues simultaneously;
• impersonation of law enforcement and government officials;
• creation of synthetic identities and bypassing KYC checks in banks, fintech services, and cryptocurrency platforms;
• Deepfake-as-a-Service, where relevant tools are sold as a ready-made criminal service.

TRM Labs identified suppliers offering individual models for face replacement and voice cloning for about $500, while annual access to a deepfake toolkit costs around $3000.

Digital forensics researcher Hany Farid from the University of California, Berkeley, demonstrated that creating real-time video impersonation could require just one screenshot and software costing $10.

A separate problem is the crypto industry: Farid's research showed that video identity checks on one of the major cryptocurrency exchanges, if conducted without human involvement, could be bypassed using synthetic video.

How to recognize a deepfake

According to Farid, people identify deepfakes only slightly better than random guessing, with overconfidence in their assessment often linked to lower accuracy. The FBI recommends paying attention to simple visual, audio, and behavioral signs:

• unnatural blinking, facial or mouth movements;
• distorted hands, accessories, shadows, and lighting;
• desynchronization of voice and lip movements;
• unnatural intonation or speech tempo;
• demands for immediate funds transfer or provision of confidential data;
• requests to switch to an unusual communication channel.

For identity verification, the FBI advises using an independent channel that is not related to the initial message. For families, a secret word can be agreed upon in advance, while in business, a callback to a known number or a second employee can be arranged to confirm financial transactions.