Deepfake fraud in 2026 is becoming infrastructure for impersonation rather than a single category of fake video.
The clearest evidence comes from systems that record different stages of the fraud process. The FBI counted $893.35 million in adjusted losses from US complaints that referenced AI. Shufti projects a 495% increase in deepfake identity fraud from its first five months of 2026. Chainalysis found that crypto scam operations linked on-chain to AI vendors generated 4.5 times more revenue per operation. Veriff and Kantar found that people in three countries performed only slightly above chance when asked to identify manipulated visuals.
These datasets cannot be combined into one global total. They count complaints, verification attempts, blockchain flows, survey responses, and controlled judgments. Read together, they show how generative AI is reducing the cost of creating identities, scaling conversations, bypassing remote checks, and moving more victims toward payment.
This page compiles more than 60 sourced AI scam and deepfake fraud data points for 2026. Government reports, original telemetry, transparent surveys, academic research, and primary regulatory material receive priority. Earlier figures remain only where they provide the latest category benchmark, a historical baseline, or a documented incident.
AI scam and deepfake fraud key numbers for 2026
How much money is connected to AI-enabled fraud?
- $893,346,472 in adjusted losses from 22,364 US complaints that reported AI-related information during 2025. The FBI published the data in April 2026.
- $632,041,188 of those losses came from investment complaints with a reported AI nexus.
- $30,256,592 came from business email compromise complaints that referenced AI.
- $19,457,078 came from tech and customer-support complaints that referenced AI.
- $19,041,653 came from confidence and romance complaints that referenced AI.
- $12,550,185 came from employment complaints that referenced AI.
- At least $14 billion reached identified crypto scam addresses during 2025. Chainalysis projects the eventual total could exceed $17 billion as more addresses are identified.
- $2.1 billion in reported US consumer losses came from scams that started on social media during 2025, according to the FTC.
- About $25 million was lost by Arup in one of the largest publicly reported corporate deepfake fraud cases.
How fast is deepfake fraud growing?
- 495% projected growth in Shufti’s deepfake identity fraud index for 2026, annualised from January to May.
- 3,892% projected growth in document deepfakes in the same dataset.
- 1,210% growth in AI or non-live fraud during 2025 in Pindrop customer data.
- 1,151% growth in iOS-targeted injection attacks during the second half of 2025 compared with the same period a year earlier, according to iProov.
- 1,400% year-over-year growth in crypto impersonation scams during 2025, according to Chainalysis.
- 58% growth in deepfake selfie attempts during 2025, according to Entrust.
- 40% year-over-year growth in injection attacks in Entrust’s 2026 report.
- More than 160,000 fraudulent verification attempts traced to only 100 facial identities in one month by Smile ID.
How much money is connected to AI-enabled fraud
The FBI’s 2025 Internet Crime Report contains the strongest current US loss table with a defined complaint-based methodology. IC3 received 1,008,597 complaints across all cybercrime categories and recorded $20.877 billion in adjusted losses. Within that total, 22,364 complaints included AI-related information and represented $893.35 million in losses.
The FBI also recorded more than $5 million in distress-scam losses. These schemes can use cloned family voices or altered proof-of-life media. The agency warns that the AI descriptor can overlap with a crime category, so the $893.35 million total should be treated as a reporting floor for complaints that explicitly mentioned AI rather than a deepfake-only estimate.
Business impact extends beyond money transferred to criminals. Veriff surveyed 1,200 fraud and compliance decision-makers in the United States and United Kingdom during 2026. Seventy-four percent reported more online fraud, 75% saw more AI-enabled attacks, and 78% expected the threat to grow during the year. Eighty-five percent said fraud had hurt revenue, including 16% who reported reductions of up to 20%.
The World Economic Forum’s Global Cybersecurity Outlook 2026 found that cyber-enabled fraud and phishing had overtaken ransomware as the leading cyber concern among chief executives. Seventy-three percent of respondents said they or someone in their network had been affected by cyber-enabled fraud during 2025.
Measurement note:
The FBI numbers measure US victim complaints. They do not estimate unreported losses, count every use of AI, or isolate deepfake audio and video from generated text, profiles, and documents.
Why there is no credible global deepfake-loss total
Search results often repeat one worldwide loss estimate without explaining what it counts. The leading datasets observe different stages of the fraud system.
Adding these figures would double-count some incidents and omit others. A single global total would imply a level of precision the evidence does not support.
The lack of a universal number is itself a useful finding. Deepfake measurement remains fragmented across law enforcement, payments, identity verification, content platforms, and research laboratories. The field needs shared reporting categories before it can support a defensible global loss estimate.
Deepfake fraud is moving into the identity stack
Shufti analysed more than one million fraud attempts and divided AI-enabled identity fraud into four categories. Its 2026 figures cover January to May and are annualised, which means they are estimates rather than completed-year totals.
Document deepfakes begin from the smallest 2025 share but carry the largest projected increase. The pattern suggests that synthetic evidence is expanding around the face. A fabricated identity can arrive with a plausible document, a matching selfie, a live stream, and a consistent backstory.
A presentation attack places a fake in front of a real camera. An injection attack bypasses the expected capture path and feeds manipulated media directly into the verification flow. iProov recorded a 1,151% increase in iOS-targeted injection attacks during the second half of 2025 and a 741% increase across the year. Entrust reported a broader 40% year-over-year rise in injection attacks.
The unit of trust is shifting from the image to the capture event. A realistic face can be synthetic, while a genuine face can arrive through a manipulated channel. Strong controls need to evaluate the person, device, camera path, behaviour, and transaction together.
Deepfake attack growth by channel
Voice and real-time conversations
- Pindrop reported a 1,300% year-over-year increase in deepfake fraud exposure during 2024 across contact-centre environments.
- The same research estimated an average $343,000 in deepfake fraud exposure per contact centre.
- Pindrop recorded 1,210% growth in AI or non-live fraud during 2025, compared with 195% growth in non-AI fraud.
- CrowdStrike recorded a 442% increase in voice phishing between the first and second halves of 2024. This figure comes from the 2025 Global Threat Report and is retained as the latest defined vishing benchmark.
- McAfee’s 2026 survey found one in ten Americans had already experienced a voice-clone scam.
Selfies, face swaps, and biometric attacks
- Entrust reported that deepfakes now account for one in five biometric fraud attempts in its dataset.
- Deepfake selfie attempts increased 58% during 2025.
- Entrust analysed more than one billion identity verifications across 195 countries and more than 30 industries.
- iProov reported that deepfake impersonation is moving beyond onboarding into routine enterprise video workflows.
- Smile ID found that nearly nine in ten South African verification attempts rejected for potential fraud involved no-face-match or spoofing signals during biometric verification.
Injection and automated attacks
- iProov reported a 1,151% increase in iOS-targeted injection attacks during the second half of 2025 and 741% growth across the full year.
- Pindrop reported 15,000 or more fraudulent bot calls against one healthcare provider during summer 2025.
- Pindrop customer data showed a 330% increase in retail AI fraud across a 60-day period in late 2025.
- Smile ID recorded more than 100,000 injection-style attempts per month across its African verification network.
These figures use different providers, periods, and denominators. They show where attacks are growing but should not be placed on one shared ranking axis.
Deepfake fraud by industry
The table separates the measured metric, observation window, and provider scope so readers can cite a row without implying that every result is directly comparable.
The largest percentage in this table does not identify the industry with the largest absolute loss. The rows measure different phenomena and provider populations. Their value is sector-specific evidence rather than a universal ranking.
Voice cloning scams
How much audio does it take to clone a voice
McAfee’s global 2023 study found that a voice clone could be created from about three seconds of audio and reach an 85% match. This remains a widely cited technical baseline, but it should be labelled as historical context rather than presented as a new 2026 result.
The same study surveyed 7,000 people. One in four had experienced a voice-cloning scam or knew someone who had. One in ten had received a cloned-voice message, and 77% of that targeted group reported losing money. Among those who lost money, 36% lost between $500 and $3,000, while 7% lost between $5,000 and $15,000.
McAfee’s 2026 State of the Scamiverse found that one in ten Americans had experienced a voice-clone scam and 35% were not confident they could identify a deepfake scam.
Trend Micro reported in April 2026 that packaged scam operations combining voice cloning, deepfake video, fake websites, and targeted messaging could be assembled for as little as $60 per month.
Concern is rising faster than protective behaviour
Malwarebytes surveyed 1,500 adults across the United States, United Kingdom, Austria, Germany, and Switzerland in 2026.
- 67% worried about voice cloning.
- Only 19% had disabled voicemail recordings.
- 81% of parents feared that a child’s likeness would be stolen.
- Only 13% had created a family codeword.
- 84% said convincing video no longer felt like proof.
- 85% said it was difficult to distinguish a scam from the real thing.
Practical implication:
Voice and video verification should rely on a known phone number, a family codeword, a second approver, or another channel the attacker did not initiate. The objective is to interrupt the request before money or credentials move.
Can people and machines detect deepfakes
Human and machine results must be shown in separate panels because the studies use different scoring systems, samples, and tasks.
Human visual detection
Veriff and Kantar surveyed 3,000 adults aged 18 to 64, with 1,000 respondents each in the United Kingdom, United States, and Brazil. Participants evaluated 16 randomised visuals.
- The United Kingdom and United States averaged 0.07 on a scale from minus one to one where zero represents random guessing.
- Brazil averaged 0.08 on the same scale.
- 70% of UK respondents misidentified a female deepfake video as genuine.
- UK awareness of the term deepfake reached 74%, but awareness did not translate into higher detection performance.
- 19% of UK respondents performed below chance and 15% scored at chance.
These values are scoring-index results rather than accuracy percentages.
Human and machine audio detection
A 2026 academic study collected 35,532 judgments from 1,768 participants across 138 text-to-speech and voice-conversion systems.
Commercial and autoregressive language-model voices were the hardest for participants to detect, with human accuracy between 61.3% and 65.9%. The reference detector maintained more than 94.5% accuracy across the study conditions, although the authors note that detectors can generalise poorly to unseen attacks.
The decline in genuine-audio recognition matters. People did not become dramatically worse at identifying fake clips. They became more likely to reject authentic speech.
Deepfakes can create harm even when a target rejects the synthetic media. A persistent stream of plausible fakes makes genuine evidence easier to dismiss. Fraud defence therefore needs to protect the provenance of authentic communication as well as detect generated content.
Why laboratory scores are incomplete
The NTIRE 2026 Robust Deepfake Detection Challenge tested image detectors against unknown common and uncommon degradations. It attracted 337 participants and 57 final submissions. The RADAR 2026 audio challenge tested more than 100,000 utterances across six languages after compression, resampling, noise, and reverberation.
A detector claim is incomplete without the dataset, media type, unseen-generator performance, false-positive rate, compression testing, and language coverage.
Selected publicly reported deepfake fraud incidents
The failed WPP and Ferrari attempts are as useful as the Arup case. Both were stopped by process rather than an employee proving that the audio or video was synthetic. Suspicion, a personal verification question, and an independent channel broke the attack chain.
Arup’s chief information officer later told the World Economic Forum that he created a basic real-time deepfake of himself in about 45 minutes using free tools. The result did not need cinematic quality to demonstrate how quickly the barrier to entry had fallen.
The economics of AI-enabled crypto scams
Chainalysis identified at least $14 billion in on-chain inflows to crypto scams during 2025 and projects that the total could exceed $17 billion as investigators identify more addresses. The average scam payment increased 253% from $782 in 2024 to $2,764 in 2025. Impersonation scam inflows grew more than 1,400% year over year, while the average severity of payments to those clusters increased by more than 600%.
The clearest AI comparison comes from operations with visible on-chain links to vendors selling face-swap software, deepfake tools, and language models.
Seventy-six percent of the AI-linked operations appeared in Chainalysis’s time-weighted high-value and high-volume quadrant. The data supports an operational interpretation. AI allows more victim conversations, identities, scripts, and channels to run in parallel.
The revenue advantage is not evidence that every synthetic face is more persuasive. It shows that generative tools improve throughput across the operation. The same scam funnel can be localised, repeated, and managed across more targets at the same time.
Related Memeburn coverage
- AI voice cloning scams and safety steps for South Africans
- Google’s case against an AI-powered phishing network
- EU complaints over financial scam ads
- How modern banking fraud is changing in South Africa
A deepfake is one asset inside a larger scam funnel
Bitdefender mapped a coordinated investment-scam network operating across at least 25 countries on six continents during 2026. The campaigns began with paid social ads, imitated trusted media and public figures, used redirect chains, and moved victims into fake news or registration pages before requesting deposits.
Counting files does not measure fraud impact. The same synthetic asset can appear at several stages, while the loss often happens after a sequence of ordinary-looking interactions. Distribution, identity consistency, and payment infrastructure can matter more than visual quality.
Deepfake fraud by geography and the South Africa picture
The most consistent country comparison in the public data remains Sumsub’s Q1 2023 to Q1 2024 benchmark. It is historical context rather than a 2026 prevalence ranking.
Current operational data adds more detail for Africa.
- Smile ID’s 2026 report covers more than 200 million identity checks across 37 industries and more than 35 African countries during 2025.
- More than 160,000 fraudulent verification attempts in one month were traced to 100 facial identities.
- Some faces appeared more than 12,000 times across several platforms.
- One synthetic identity was attempted more than 1,000 times within 30 minutes.
- Smile ID detected 71% more duplicate fraud attempts in 2025 than in 2023 and 2024 combined.
- Southern African deepfake attempts increased from fewer than 200 per month in 2024 to more than 3,000 per month by the end of 2025.
- In South Africa, 47% of potentially fraudulent biometric verifications were rejected for no face match and 40% for spoofing.
- iProov recorded a 720% attack spike in Southeast Asia during the third quarter of 2025. This is a separate provider dataset and should not be compared directly with Sumsub’s country figures.
What is known about deepfake fraud in South Africa
The Financial Sector Conduct Authority issued a warning on 5 June 2026 about a deepfake investment scam impersonating financial journalists Maya Fisher-French and Bruce Whitfield. The warning confirms that synthetic endorsement fraud is reaching South African consumers through investment promotion.
No consolidated public dataset currently reports national South African deepfake incidents and losses across advertising, banking, identity verification, account takeover, and voice impersonation. A precise national total would therefore be misleading.
South Africa has both an active threat signal and a measurement gap. Vendor networks show rapid growth in biometric impersonation, while the regulator confirms local investment scams. Public reporting still lacks a shared taxonomy that connects those incidents to losses.
Who reports exposure and where scams begin
Strong current sources support a clearer consumer picture than celebrity-count lists from SEO statistics pages.
Exposure by age
Malwarebytes found that 50% of surveyed adults had experienced some form of AI fraud or scam.
Nineteen percent of respondents had experienced AI-driven identity harm, rising to 30% among Gen Z. One in ten reported that explicit AI images had been created of them without consent.
Complaint categories and distribution channels
FBI AI-referenced complaint data shows that harm reaches investors, workers, families, businesses, and people seeking support. The largest loss categories were investment fraud, business email compromise, tech support, confidence and romance fraud, personal data breaches, and employment scams.
FTC data shows how these schemes reach victims. In 2025, nearly 30% of people who reported losing money to a scam said it started on social media. Reported social-media scam losses reached $2.1 billion. Investment scams caused $1.1 billion of that total, more than half. Nearly 60% of reported romance-scam losses began on social media, and one in three people who lost money to a job or business-opportunity scam said it began there.
These FTC figures cover all scams initiated on social media rather than deepfakes alone. They matter because social platforms supply the targeting, advertising, fake profiles, and private conversations that synthetic media can strengthen.
What effective defence looks like in 2026
No current evidence supports relying on one detector or a person’s visual judgment. Stronger controls force an attacker to defeat several independent signals.
Google reported in May 2026 that SynthID had watermarked more than 100 billion images and videos and 60,000 years of audio. SynthID verification had been used 50 million times globally. Google is also expanding C2PA Content Credentials, which can show how media was captured or modified.
Provenance is not a complete fraud defence. It can strengthen evidence about origin, while capture integrity, identity graphs, transaction monitoring, and independent approval address the wider scam process.
Regulation and industry response
United States
The TAKE IT DOWN Act was signed in May 2025. It criminalises certain publication of nonconsensual intimate imagery, including AI-generated material. From 19 May 2026, covered platforms must provide a removal process and take down validly reported intimate content and known identical copies within 48 hours. The FTC began enforcing the platform requirements and sent warning letters to a dozen websites on 20 May 2026.
The first conviction highlighted under the law came on 9 April 2026 after an Ohio man pleaded guilty in a case involving AI-generated nonconsensual intimate images. This is a regulatory milestone rather than a financial-fraud incident.
The DEFIANCE Act passed the Senate again in January 2026 and would create a civil remedy for nonconsensual sexually explicit deepfakes. It still awaited House consideration in June 2026.
State-law counts vary according to scope. Public Citizen maintains separate trackers for election deepfakes and intimate deepfakes. A single claim that a certain number of states has deepfake laws can mix enacted measures, introduced bills, election rules, criminal provisions, civil remedies, and platform requirements. This report therefore does not collapse the trackers into one headline number.
European Union
Article 50 transparency obligations under the EU AI Act become applicable on 2 August 2026. They cover marking and detection of AI-generated content and disclosure of deepfakes and certain AI-generated publications. This date applies to the Article 50 transparency rules rather than representing the start of every obligation or enforcement provision in the AI Act.
Detection and provenance
Regulators are moving alongside technical responses. Watermarking, content credentials, capture-origin signals, and deepfake-detection services can improve transparency. Their effectiveness still depends on adoption across models, devices, platforms, and the channels where content is compressed or re-recorded.
The 2026 deepfake fraud timeline
Seven conclusions supported by the data
- The fastest reported growth is happening around the face. Documents, synthetic identities, and injection attacks are turning one generated likeness into a broader identity package.
- Deepfake fraud is a systems problem. Advertising, messaging, identity checks, payment, and laundering can all contribute to one loss. A detector at one stage cannot secure the full path.
- Human confidence is an unreliable control. Awareness did not produce strong visual-detection performance, while authentic audio is increasingly misclassified as fake.
- AI changes fraud economics. Chainalysis found higher revenue, daily income, and transaction frequency among scam operations with visible links to AI vendors.
- Authentic media needs protection. Detection alone does not address the erosion of trust in genuine communication. Provenance and independent verification become more important as plausible fakes spread.
- Benchmark context matters. A detector score needs its dataset, generator coverage, degradation tests, false-positive rate, and language scope.
- Measurement remains fragmented. Government complaints, vendor telemetry, blockchain flows, surveys, and research benchmarks provide complementary views. They should be compared without being added together.
FAQs
How fast is deepfake identity fraud growing in 2026?
Shufti projects 495% growth over 2025 based on an annualised January to May run rate. The figure is a provider-network estimate labelled 2026E rather than a completed global total.
What is the fastest-growing attack type in the Shufti data?
Document deepfakes carry a projected 3,892% year-over-year increase. They use generated or altered identity evidence to support a synthetic or stolen persona.
How much was lost in US complaints that referenced AI?
The FBI recorded $893,346,472 in adjusted losses across 22,364 complaints during 2025. The total includes several uses of AI and is not limited to deepfake audio or video.
Can people reliably identify deepfake visuals?
Veriff and Kantar found average scores of 0.07 in the United Kingdom and United States and 0.08 in Brazil on a scale where zero represents random guessing.
Are AI-linked crypto scams more profitable?
Chainalysis found average revenue of $3.2 million per operation among scams with visible on-chain links to AI vendors, compared with $719,000 among the comparison group.
What is one of the largest publicly reported corporate deepfake fraud cases?
Arup lost about $25 million after an employee joined a video meeting with synthetic versions of the chief financial officer and colleagues and made 15 transfers.
How much audio can be enough to create a voice clone?
McAfee’s 2023 technical test found that about three seconds could create a clone with an 85% match. It remains historical context rather than a new 2026 measurement.
Why do deepfake statistics disagree?
The sources count different units, including complaints, losses, blockchain flows, verification attempts, survey responses, and benchmark judgments. They can describe the same system without being mathematically compatible.
What is known about deepfake fraud in South Africa?
Smile ID recorded a steep increase in Southern African deepfake attempts and high levels of biometric impersonation signals in South Africa. The FSCA also warned about a local deepfake investment scam in June 2026. No consolidated public national loss total is available.
What deepfake rules apply in 2026?
The FTC began enforcing TAKE IT DOWN platform removal duties in May 2026. EU AI Act Article 50 transparency obligations apply from 2 August 2026. State-level US rules vary by election, intimate-image, civil, criminal, and platform scope.
References
- Federal Bureau of Investigation. (2026). 2025 IC3 Annual Report. [online] Internet Crime Complaint Center. Available at: https://www.ic3.gov/AnnualReport/Reports/2025_IC3Report.pdf [Accessed 1 July 2026].
- Shufti Pro. (2026). Deepfake Fraud Set to Surge 495% in 2026. [online] Shufti Pro. Available at: https://shuftipro.com/resources/whitepapers-reports/deepfake-identity-fraud-index-report-2026/ [Accessed 1 July 2026].
- Chainalysis. (2026). 2026 Crypto Crime Report on Scams. [online] Chainalysis. Available at: https://www.chainalysis.com/blog/crypto-scams-2026/ [Accessed 1 July 2026].
- World Economic Forum. (2026). Global Cybersecurity Outlook 2026. [online] World Economic Forum. Available at: https://www.weforum.org/publications/global-cybersecurity-outlook-2026/ [Accessed 1 July 2026].
- Veriff. (2026). Fraud Industry Pulse Survey 2026. [online] Veriff. Available at: https://www.veriff.com/pr-news/ai-fraud [Accessed 1 July 2026].
- Veriff and Kantar. (2026). Deepfakes Report 2026 UK. [online] Veriff. Available at: https://www.veriff.com/pr-news/deepfakes-report-2026-uk [Accessed 1 July 2026].
- Entrust. (2026). 2026 Identity Fraud Report. [online] Entrust. Available at: https://www.entrust.com/resources/reports/identity-fraud-report [Accessed 1 July 2026].
- iProov. (2026). 2026 Threat Intelligence Report. [online] iProov. Available at: https://www.iproov.com/press/iproov-issues-annual-threat-intelligence-report [Accessed 1 July 2026].
- Pindrop. (2026). Inside the 2025 AI Fraud Spike. [online] Pindrop. Available at: https://www.pindrop.com/ai-fraud-spike/ [Accessed 1 July 2026].
- Pindrop. (2026). Deepfake Fraud and Business Exposure. [online] Pindrop. Available at: https://www.pindrop.com/article/defending-businesses-with-deepfake-detection/ [Accessed 1 July 2026].
- CrowdStrike. (2025). 2025 Global Threat Report Findings. [online] CrowdStrike. Available at: https://www.crowdstrike.com/en-us/blog/crowdstrike-2025-global-threat-report-findings/ [Accessed 1 July 2026].
- Malwarebytes. (2026). Face Value Research on AI, Trust, Scams and Identity Theft. [online] Malwarebytes. Available at: https://www.malwarebytes.com/press/2026/06/10/new-malwarebytes-research-reveals-how-ai-is-reshaping-trust-scams-and-identity-theft [Accessed 1 July 2026].
- Smile ID. (2026). 2026 Digital Identity Fraud in Africa Report. [online] Smile ID. Available at: https://smile.id/blog/2026-digital-identity-fraud-in-africa-report-ai-driving-unprecedented-fraud-scale-and-sophistication-successful-firms-treat-identity-as-a-security-surface [Accessed 1 July 2026].
- Sumsub. (2024). Deepfake Cases Surge in Countries Holding 2024 Elections. [online] Sumsub. Available at: https://sumsub.com/newsroom/deepfake-cases-surge-in-countries-holding-2024-elections-sumsub-research-shows/ [Accessed 1 July 2026].
- McAfee. (2026). 2026 State of the Scamiverse. [online] McAfee. Available at: https://www.mcafee.com/blogs/wp-content/uploads/2026/01/Scamiverse.pdf [Accessed 1 July 2026].
- McAfee. (2026). Global AI Voice-Cloning Study. [online] Memeburn. Available at: https://memeburn.com/wp-admin/post.php?post=227981&action=edit [Accessed 1 July 2026].
- Trend Micro. (2026). AI Voice Cloning and Scam Assembly Lines. [online] TrendLife. Available at: https://trendlife.com/en-us/blog/2026/04/16/ai-voice-cloning [Accessed 1 July 2026].
- Müller, N.M. and Choong, W.H. (2026). Eroding Trust in Real Speech. [online] arXiv. Available at: https://arxiv.org/abs/2605.26136 [Accessed 1 July 2026].
- Hopf, B. et al. (2026). Robust Deepfake Detection, NTIRE 2026 Challenge Report. [online] IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops. Available at: https://openaccess.thecvf.com/content/CVPR2026W/NTIRE/html/Hopf_Robust_Deepfake_Detection_NTIRE_2026_Challenge_Report_CVPRW_2026_paper.html [Accessed 1 July 2026].
- Luong, H.T., Liu, X., Kukanov, I., Chai, Z.X. and Lee, K.A. (2026). RADAR Challenge 2026: Robust Audio Deepfake Recognition under Media Transformations. [online] arXiv. Available at: https://arxiv.org/abs/2605.09568 [Accessed 1 July 2026].
- Bitdefender Labs. (2026). Global Investment Scam Network Using Meta Ads. [online] Bitdefender. Available at: https://www.bitdefender.com/en-us/blog/labs/global-investment-scam-network-using-meta-ads [Accessed 1 July 2026].
- World Economic Forum. (2025). Arup Deepfake Fraud Case. [online] World Economic Forum. Available at: https://www.weforum.org/stories/2025/02/deepfake-ai-cybersecurity-fraud/ [Accessed 1 July 2026].
- The Guardian. (2024). WPP Chief Targeted in Deepfake Scam Attempt. [online] The Guardian. Available at: https://www.theguardian.com/technology/article/2024/may/10/wpp-deepfake-scam-mark-read [Accessed 1 July 2026].
- The Guardian. (2024). Ferrari Executive Foils Deepfake Scam by Asking CEO a Question Only He Could Answer. [online] The Guardian. Available at: https://www.theguardian.com/technology/article/2024/jul/26/ferrari-executive-foils-deepfake-scam-by-asking-ceo-a-question-only-he-could-answer [Accessed 1 July 2026].
- Financial Sector Conduct Authority. (2026). Warning about a Deepfake Investment Scam Involving Maya Fisher-French and Bruce Whitfield. [online] Financial Sector Conduct Authority. Available at: https://www.fsca.co.za/ [Accessed 1 July 2026].
- Federal Trade Commission. (2026). Reported Losses to Scams on Social Media Eight Times Higher Than in 2020. [online] Federal Trade Commission. Available at: https://www.ftc.gov/news-events/data-visualizations/data-spotlight/2026/04/reported-losses-scams-social-media-eight-times-higher-2020 [Accessed 1 July 2026].
- Federal Trade Commission. (2026). FTC Begins Enforcing the TAKE IT DOWN Act. [online] Federal Trade Commission. Available at: https://www.ftc.gov/news-events/news/press-releases/2026/05/ftc-begins-enforcing-take-it-down-act [Accessed 1 July 2026].
- U.S. Senate Committee on Commerce, Science, and Transportation. (2026). TAKE IT DOWN Act Leads to Conviction in Case Targeting AI-Generated Deepfakes. [online] U.S. Senate Committee on Commerce, Science, and Transportation. Available at: https://www.commerce.senate.gov/press/rep/release/cruz-klobuchar-take-it-down-act-leads-to-conviction-in-case-targeting-ai-generated-deepfakes/ [Accessed 1 July 2026].
- Office of U.S. Senator Dick Durbin. (2026). Durbin Questions Witnesses during Senate Judiciary Subcommittee Hearing on Protecting Kids Online. [online] U.S. Senator Dick Durbin of Illinois. Available at: https://www.durbin.senate.gov/newsroom/press-releases/durbin-questions-witnesses-during-senate-judiciary-subcommittee-hearing-on-protecting-kids-online [Accessed 1 July 2026].
- Public Citizen. (2026). State Legislation on Deepfakes in Elections. [online] Public Citizen. Available at: https://www.citizen.org/article/tracker-legislation-on-deepfakes-in-elections/ [Accessed 1 July 2026].
- Public Citizen. (2026). State Legislation on Intimate Deepfakes. [online] Public Citizen. Available at: https://www.citizen.org/article/tracker-intimate-deepfakes-state-legislation/ [Accessed 1 July 2026].
- European Commission. (2026). Code of Practice on Transparency of AI-Generated Content. [online] Shaping Europe’s Digital Future. Available at: https://digital-strategy.ec.europa.eu/en/policies/code-practice-ai-generated-content [Accessed 1 July 2026].
- European Commission. (2026). Commission Opens Consultation on Draft Guidelines on AI Transparency Obligations. [online] Shaping Europe’s Digital Future. Available at: https://digital-strategy.ec.europa.eu/en/news/commission-opens-consultation-draft-guidelines-ai-transparency-obligations [Accessed 1 July 2026].
- Google. (2026). Making It Easier to Understand How Content Was Created and Edited. [online] Google. Available at: https://blog.google/innovation-and-ai/products/identifying-ai-generated-media-online/ [Accessed 1 July 2026].
The post Deepfake Statistics 2026 Reveal a 3,892% Fraud Surge appeared first on Memeburn.