digital democracyartificial intelligence

Deepfakes cost billions as incidents hit 114 countries

Deepfakes — AI-generated videos, images, or audio that show people saying or doing things they never did — have quickly become a powerful tool for scammers and other bad actors. Our research reveals how fast deepfake incidents are growing, which countries are most affected, and which scams and platforms are driving billions in financial losses from deepfake fraud.

Deepfake incidents have surged in recent years

For this research, we analyzed deepfake-related incidents involving the creation of fake videos, images, or audio that were covered in news articles. We compiled the incident data from three sources: the AI Incident Database, Resemble AI, and the OECD (Organisation for Economic Co-operation and Development).

In total, we identified 2,430 incidents between 2020 and the third quarter of 2026. Notably, nearly 90% of all incidents occurred from 2024 onwards, highlighting how heavily reported deepfake incidents are concentrated in the last few years, with recent deepfake incidents in 2025 and 2026 making up the bulk of cases. Early volumes were relatively low, with 22 incidents in 2020, 19 in 2021, and 41 in 2022. However, 2023 marked the first major surge, with incidents more than quadrupling to 170. This was followed by 399 in 2024 and the largest annual increase, when deepfake incidents in 2025 reached 970.

With 809 incidents already documented in the first three quarters of 2026, this year is on pace to set a new record, representing a 24% increase compared to the same period in 2025 (651 incidents). Furthermore, every single quarter in 2026 recorded higher volumes than in 2025, with increases of 47% in the first quarter, 23% in the second, and 3% in the third.

Deepfake-related incidents were recorded in 114 countries. The top eight countries by incident volume were spread across every inhabited continent except Africa. The United States led with 787 incidents, more than three times as many as India, which ranked second with 246. The United Kingdom came third with 141, followed by China (111), Brazil (79), South Korea (76), Taiwan (62), and Australia (62).

North America had the most incidents (871), followed closely by Asia (853). Europe accounted for 620 incidents, South America for 132, Oceania for 78, and Africa for 64. Some incidents involved multiple continents and are therefore counted more than once in this breakdown. Overall, these figures show that deepfake-related incidents are a global phenomenon, with cases occurring across every inhabited continent.

Deepfake-enabled fraud caused an estimated $3.91 billion in financial losses globally

The $3.91 billion in financial losses from deepfake-related fraud spanned 51 countries in our research, with 98% of the total occurring from 2024 onwards. Between 2020 and 2023, combined reported losses totaled $83 million. The scale of deepfake financial fraud escalated rapidly in 2024, when losses jumped to $335 million, then reached a record $2.52 billion in 2025 — 7.5 times higher than the year before. Through the first three quarters of 2026, losses have already reached $972 million.

Looking at the most recent 12-month period (2025 Q4 through 2026 Q3), financial losses reached $2.24 billion, representing a 55% year-over-year increase compared to the $1.44 billion recorded across the previous four quarters (2024 Q4 through 2025 Q3). This surge was driven in large part by a record-breaking fourth quarter in 2025, which alone accounted for $1.26 billion in losses, followed by $452 million in Q1 2026, $315 million in Q2 2026, and $206 million in Q3 2026.

The United States recorded the highest financial losses globally at $705 million, five times more than neighboring Canada in sixth place ($141 million). Two Southeast Asian countries followed closely behind: Indonesia ranked second ($615 million) and Malaysia ranked third ($513 million) — meaning their combined losses exceeded the US total by 60%. Rounding out the top five were Europe's hardest-hit nations, the United Kingdom ($177 million) and Italy ($148 million).

At the continental level, Asia led with $1.61 billion in losses, followed by Europe ($1.34 billion) and North America ($847 million), while Oceania ($23 million), Africa ($16.5 million), and South America ($4.6 million) saw much smaller figures. Europe’s high total was also spread across more countries: 14 European countries recorded losses exceeding $50 million, compared to seven in Asia and just two in North America. An additional $65 million in losses could not be attributed to a specific continent and is not included in this breakdown.

Investment scams lead deepfake fraud, accounting for 45% of all financial losses

Fraud-related deepfakes accounted for 768 out of 2,430 incidents in our research. Investment scams, in which criminals use deepfakes of celebrities and public figures to promote fraudulent investment schemes, ranked first in both incident volume and financial damage. These scams caused $1.75 billion in losses, accounting for 45% of the $3.91 billion total, and made up 33% of all recorded deepfake fraud incidents (254 of 768).

Finance and banking fraud ranked second in financial losses at $1.14 billion (29% of the total) and third in incident volume, accounting for 13% of cases (103 of 768). In our study, this category of AI deepfake banking fraud includes bank impersonation, phishing, voice-cloned payment requests, and deepfakes used to bypass identity checks or authorize transfers.

Corporate fraud, in which scammers impersonate CEOs, other employees, or business partners to request fraudulent transactions or commit other business-related fraud, ranked third in financial losses at $545 million (14% of the total) and fourth in incident volume, accounting for 7% of cases (56 of 768).

Product promotion scams, in which deepfakes are used to endorse fake, counterfeit, or non-existent goods, ranked fourth in financial losses at $277 million (7% of the total) and second in incident volume, representing 18% of cases (140 of 768).

Rounding out the top five were romance scams, where criminals use synthetic videos, images, or voice clones to deceive victims in fake online relationships, accounting for $179 million in losses (5% of the total) across 30 incidents (4% of cases).

The remaining 185 incidents (24% of fraud cases) fell into other categories, accounting for $14.6 million in losses (less than 1% of the total). These cases included impersonating family members or government agencies, extorting victims with fabricated explicit images, and even sending fake photos of lost pets to demand money, among other schemes. While these incidents resulted in smaller financial losses, they still likely caused stress and emotional harm to victims.

Social media drives the majority of deepfake fraud

Social media was the primary first-contact channel for spreading deepfake-related fraud, accounting for 57% of recorded cases (438 of 768) and $2.16 billion in financial losses (55% of the total). An additional 65 incidents (8% of cases) and $765 million in losses (20%) were attributed to websites. Because news reports in this category often used broad terms like ‘’online ads,’’ many of these schemes likely also spread through social media.

Phone calls ranked second in incident volume at 12% of cases (93 incidents) and fifth in financial losses at $181 million (5% of the total), with scammers using AI audio to impersonate family members, police officers, or company employees. Video calls took this deception a step further by deploying real-time deepfakes during live conversations. Although video calls accounted for just 3% of cases (25 incidents), they caused $363 million in losses (9% of the total), more than double those from phone calls, primarily by targeting companies in live meetings or deceiving victims in romance scams.

Authentication and biometric bypass, a form of deepfake identity fraud where scammers used AI-generated images or videos to pass identity verification checks, access victims' bank accounts, or take out loans in their names, accounted for 3% of cases (25 incidents) and $276 million in losses (7% of the total).

Meanwhile, messaging apps such as Telegram and WhatsApp accounted for 6% of cases (46 incidents) and $84 million in financial losses (2% of the total). On these platforms, scammers primarily used AI voice notes and videos to impersonate company executives or family members, lure victims into fake investment groups, and conduct sextortion or romance scams. Although these figures represent only cases where messaging apps were the primary channel, these platforms are also frequently used as a secondary step in social media scams, as criminals move victims to private chats to complete the fraud.

When examining which social media platforms served as first-contact channels in deepfake-related fraud, 267 out of 438 incidents named a specific platform. Among those cases, Facebook was the most frequently mentioned at 42%, followed by Instagram at 25%, X (formerly Twitter) and YouTube at 13% each, and TikTok at 12%, while other social media platforms appeared in 9% of cases. In some instances, multiple platforms were involved in a single incident.

A similar breakdown of financial losses by first-contact platforms shows an even steeper concentration. Although only $846 million (39%) of the $2.16 billion in social media losses could be attributed to a named platform, the available data is heavily concentrated on Meta-owned networks, with Facebook accounting for $679 million (80%) and Instagram for $164 million (19%). In comparison, all other platforms combined totaled $2.2 million (less than 1%). While unreported platform details in news coverage make it difficult to draw definitive conclusions, the trend indicates that Facebook and Instagram are the primary launchpads for high-loss deepfake scams.

How to spot a deepfake

As deepfakes become more realistic and widespread, recognizing them takes more effort than ever. Tomas Stamulis, Chief Security Officer at Surfshark, explains what to look out for to avoid deepfake fraud:

Detection tools often struggle to keep up with deepfake technology, so spotting a fake now takes more than a quick glance. Look closely at the details: unnatural movements, mismatched colors or lighting, distorted backgrounds, and lips that don’t quite match the audio can all be warning signs. It’s also worth noting how the content reached you, as deepfakes are often spread by bot accounts.

Stamulis adds that context is just as important:

As these flaws become harder to see, be cautious with any message that creates urgency, asks for money, or comes from someone contacting you through unusual channels, as these are common signs of deepfake financial fraud. Seeing or hearing someone is no longer enough to know it’s really them. So when in doubt, pause and verify through a channel you trust.

Methodology and sources

This study used data from the AI Incident Database, Resemble AI, and the OECD to create a combined dataset covering deepfake incidents from January 2020 to September 2026. Incidents were included if they referenced synthetic videos, images, or audio and were verified by media reports. The first section of this study analyzes global trends across all 2,430 recorded deepfake incidents, while the subsequent sections focus specifically on the 768 incidents involving financial fraud and scams.

Each incident was classified by the affected country and continent. While each multi-country incident was counted only once in the global total (2,430), it was counted in each affected location for country and continent volume breakdowns. However, for financial loss breakdowns, if an incident with a reported financial loss involved multiple countries, the monetary loss was divided equally among the countries involved. Reported financial losses in non-US currencies were converted to US dollars (USD). Incidents where a specific country or continent could not be identified were included in global totals but excluded from geographic breakdowns.

Each fraud incident was assigned to one primary first-contact channel based on the main platform, tactic, or entry point described in the source material:

• Social media: incidents where victims first encountered the deepfake or scam via social networks (such as Facebook, Instagram, X/Twitter, YouTube, or TikTok). When news reports listed multiple social media platforms as first-contact channels for a single incident, each named platform was recorded in the platform-specific breakdown;
• Websites: incidents originating on standalone websites, fake news pages, or broadly described "online ads" where a specific social network was not named;
• Phone calls: scams initiated via traditional telephone calls using AI voice cloning or synthetic audio;
• Video calls: incidents in which scammers used real-time or pre-recorded deepfake video during live video meetings or calls;
• Messaging apps: scams in which initial contact took place on messaging platforms such as WhatsApp, Telegram, Signal, and Viber;
• Authentication and biometric bypass: cases where AI-generated images or videos were used directly to trick identity verification or facial recognition systems to access accounts or take out loans.

Data was collected from: