What is the total cost of deepfake fraud globally?

Comprehensive global figures are not yet available because most incidents are not categorized as specifically involving deepfakes. The FBI IC3 2024 report attributed $12.5 billion to voice phishing specifically. Industry analysts estimate total deepfake-assisted fraud across identity, financial, and romance categories at $40–60 billion annually by 2026, though this figure includes significant estimation uncertainty.

How fast is deepfake fraud growing?

Deepfake-assisted KYC fraud attempts grew 340% year-over-year in 2025 according to Deloitte. Voice phishing attack volume doubled year-over-year in 2025. The overall trend is accelerating as generation tools become cheaper, faster, and more accessible to non-technical fraudsters.

What percentage of deepfakes are detected?

Without systematic detection in place, organizations detect very few deepfakes. Human detection rates on modern deepfakes approach random chance. With AI-based detection like ScamAI, organizations typically find that 1–5% of submitted KYC media contains detectable deepfake characteristics — a figure that varies significantly by industry, geography, and the population being screened.

Are deepfakes getting more or less detectable over time?

Both. Generation quality is improving, making deepfakes harder to detect visually. Simultaneously, detection technology is improving, with models trained on the latest generation tools achieving higher accuracy than earlier detectors. The net result is that dedicated AI detection remains effective, while human detection continues to degrade.

What is the regulatory landscape for deepfake detection?

Regulation is evolving rapidly and varies by jurisdiction. The EU AI Act requires labeling of AI-generated content. The U.S. DEFIANCE Act creates federal liability for intimate deepfakes. Multiple countries have enacted criminal penalties for deepfake fraud. Organizations handling identity verification should monitor their specific regulatory environment, as mandatory detection requirements are increasingly likely.

Deepfake statistics 2026: the numbers you need to know

The growth of deepfake creation

AI-generated media has gone from a research curiosity to a mass consumer behavior in under five years. The key driver is the commoditization of generation tools: creating a convincing AI-generated image went from requiring weeks of GPU time on specialized hardware in 2019 to requiring 30 seconds on a free consumer platform in 2026.

3,000% increase in AI-generated media circulating online between 2022 and 2026 (ScamAI Research)
Over 500 million AI-generated images are estimated to be created every day across major platforms as of 2026
Voice cloning models can produce a convincing clone from under 30 seconds of source audio — down from hours of source material required in 2020
GPT-Image-2 generated over 1 million user-reported AI images on Twitter within its first week of deployment (arXiv:2604.25370)
47 documented state-actor influence campaigns using AI-generated media were identified in 2025

Financial fraud losses attributed to deepfakes

Key Stat

$12.5 billion in global losses attributed to voice phishing attacks in 2024, many involving AI-cloned voices — FBI Internet Crime Complaint Center (IC3) 2024 Annual Report.

$12.5 billion in losses attributed to voice phishing (vishing) globally in 2024 (FBI IC3)
340% increase in deepfake-assisted KYC fraud attempts in 2025 (Deloitte)
$25 million lost by a single multinational company in one deepfake video call attack impersonating the CFO (2024)
280% increase in AI-edited photo fraud in insurance claims between 2024 and 2025
Over 70% of reported romance scam profiles now involve AI-generated or manipulated imagery (FTC)
Synthetic identity fraud — creating fake identities using AI-generated documents and photos — accounts for an estimated $6 billion in annual credit losses (Federal Reserve estimates)

These figures likely undercount the true scale of deepfake fraud. Many incidents go unreported, and organizations frequently cannot determine whether fraud involved deepfakes specifically without forensic analysis. As detection becomes more widespread, the proportion of fraud attributed to synthetic media is expected to increase as more incidents are correctly categorized.

Detection accuracy benchmarks

Detection accuracy statistics reveal a significant gap between claimed performance and real-world performance. Most academic and commercial benchmarks test detectors on controlled datasets that overlap with training data — producing optimistic accuracy figures that do not hold in production.

ScamAI Eva-v1 model: 95.3% accuracy on in-the-wild image deepfakes (ScamAI internal benchmark, 2026)
ScamAI audio model: 98.5% accuracy on voice clone detection across major synthesis platforms
Open-source detection models: 50–65% accuracy on out-of-distribution deepfakes from tools not in training data (arXiv:2602.07814)
Human reviewers: accuracy approaching random chance (50%) on high-quality modern deepfakes
Human reviewers with training: performance improves to approximately 60–65% — still significantly below AI-based detection
GPT-Image-2 detecting its own generated documents: failed at rates consistent with random chance, confirming that generation tools cannot self-detect (arXiv:2604.25213)

Key Stat

Human detection of high-quality deepfakes performs near random chance (50%), compared to ScamAI's 95.3% AI-based detection. This gap is the primary argument for automated detection systems.

Industry-specific impact statistics

Deepfake fraud is not uniformly distributed across industries. Financial services, identity verification, and media organizations face the highest incident rates due to the high-value targets they represent.

Financial services — 1 in 7 digital onboarding fraud attempts now involves a deepfake element (ScamAI detection data, 2026)
Dating platforms — over 70% of romance scam profiles involve AI-generated profile photos (FTC 2025)
Insurance — 280% increase in AI-edited claims photos between 2024 and 2025
Call centers — vishing attack volume doubled year-over-year in 2025, with AI-cloned voices cited in the majority of sophisticated attacks
Media and journalism — the number of deepfake political videos circulating before elections doubled in 2024 compared to 2022
HR and hiring — approximately 1 in 14 remote job applicants now uses AI-assisted video tools during interviews (LinkedIn data, 2025)
Government — 47 documented state-actor influence campaigns using AI-generated media in 2025 (ScamAI Research)

Regulatory and policy response

Government regulation of deepfakes has accelerated significantly since 2024. The regulatory landscape is fragmented — different jurisdictions have taken different approaches, from criminal penalties to platform liability frameworks.

United States — the DEFIANCE Act (2024) created federal civil liability for non-consensual intimate deepfakes. Over 20 U.S. states have specific deepfake legislation
European Union — the AI Act (2024) requires explicit labeling of AI-generated media and creates liability for deepfake-based fraud
United Kingdom — the Online Safety Act provisions covering intimate deepfakes took effect in 2024
South Korea — criminal penalties for deepfake distribution were strengthened following high-profile cases in 2024
China — regulations requiring watermarking of all AI-generated content took effect in 2023 and were extended in 2025
Australia — the Online Safety (Deepfake Sexual Material) Amendment passed in 2024

The regulatory trend is clearly toward mandatory detection and labeling obligations for platforms that host user-generated content, with particular focus on electoral integrity and financial fraud use cases. Organizations that implement detection now are building the compliance infrastructure that will likely be required in their jurisdiction within the next 2–3 years.

Key takeaways for security and compliance teams

The numbers point to three conclusions for organizations evaluating their synthetic media risk.

First, the fraud is already happening. A 340% year-over-year increase in deepfake KYC attempts means organizations without detection are already being defrauded — they simply cannot see it. Detection does not prevent new fraud; it makes existing fraud visible.

Second, human review cannot scale. Human accuracy at random chance means that adding headcount to manual review of media is an ineffective response. AI detection at 95.3%+ accuracy is the only approach that scales.

Third, the cost of detection is negligible relative to fraud losses. At $0.05 per image analyzed, screening 10,000 KYC submissions costs $500. The average cost of a single fraudulent account opened via deepfake-assisted KYC bypass typically exceeds $5,000 in direct losses. The ROI on detection is extremely favorable.

Pro Tip

At $0.05 per image, screening 10,000 KYC submissions costs $500. A single deepfake-assisted fraud account typically costs more than $5,000 in direct losses. Detection pays for itself with the first fraud it prevents.