Deepfake Statistics 2026: Fraud, Detection, and Spread
People spot deepfakes at about the rate of a coin flip: a review of 56 studies put human accuracy at 55.5%, not significantly better than chance. In one test of 2,000 people, only 0.1% correctly labeled every real and fake image and video. Deloitte expects generative AI to push US fraud losses to $40 billion by 2027, and one finance worker has already sent $25 million after a video call where every other face was fake. The most quoted number about scale, “8 million deepfakes shared in 2025,” started as a company blog’s extrapolation. Here are the deepfake statistics worth knowing in 2026, each sourced inline with the year noted.
Humans detect deepfakes at roughly chance
Human accuracy at detecting high-quality deepfakes is 55.5%, not significantly above chance, in a meta-analysis of 56 papers covering 86,155 participants (Diel et al., Computers in Human Behavior Reports, 2024).
Accuracy was 62% for audio, 57% for video, 53% for images and 52% for text, with wide confidence intervals (Diel et al., 2024).
Feedback training, AI support and other strategies raised accuracy to about 65%, especially for video (Diel et al., 2024).
Only 0.1% of 2,000 UK and US consumers correctly identified every real and deepfake image and video, even though they were told to look for fakes, in a study by identity verification company iProov (iProov, 2025).
Participants were 36% less likely to correctly identify a fake video than a fake image, and more than 60% were confident in their answers whether they were right or not (iProov, 2025).
22% of consumers had never heard of deepfakes, rising to 39% of those 65 and older (iProov, 2025).
$40 billion in AI-enabled fraud by 2027
Generative AI could push US fraud losses to $40 billion by 2027, up from $12.3 billion in 2023, a 32% annual growth rate, Deloitte’s Center for Financial Services forecasts (Deloitte, 2024). It is a projection, and it covers all generative AI fraud, not deepfakes alone.
A finance worker at engineering firm Arup paid out $25 million after a video call in which the CFO and other colleagues were deepfakes (CNN, 2024). The worker had doubts until the call, because everyone on it looked and sounded like colleagues the worker recognized.
Voice deepfakes rose 680% in 2024, and the company forecast a further 162% rise in deepfake fraud in 2025, by Pindrop, which sells voice fraud detection to call centers (Pindrop, 2025).
The FBI warned in May 2025 that attackers had been sending AI-generated voice messages impersonating senior US officials since April 2025, aiming to get targets onto other messaging platforms and into their accounts (FBI IC3, 2025).
“8 million deepfakes in 2025” was a blog’s extrapolation
“An estimated 8 million deepfakes were shared in 2025, up from 500,000 in 2023” was reported as government figures in February 2026, when the UK announced a deepfake detection framework with Microsoft (The Standard, 2026).
A year earlier, the UK government had called it a projection: “a projected eight million will be shared in 2025” (GOV.UK, 2025). That projection links to a blog post by software company OpenFox.
The OpenFox post took a 2023 estimate of 500,000 deepfakes on social media from DeepMedia, a deepfake detection startup, and assumed the number doubles every six months (OpenFox, 2024). Four doublings of 500,000 is 8 million. The figure is arithmetic on a startup’s estimate, not a count.
For the wider threat picture, see cybersecurity statistics and AI statistics.
Putting a check between a convincing request and the money
Every deepfake fraud above ends the same way: someone acts on a request that looks real, usually a payment or a change of bank details, without checking through a second channel. Carly is an AI assistant that works in your email and calendar. You can have her watch your inbox for payment requests, new wire instructions or “urgent” asks from executives, and flag each one for a call-back on a number you already have before anything moves. Carly connects to thousands of apps, so the check fits into the tools your finance team already uses.
For more, see how to report phishing in Outlook and email statistics.
FAQ
Can people detect deepfakes? Barely. A meta-analysis of 56 studies found humans identify deepfakes with 55.5% accuracy, not significantly better than chance (Diel et al., 2024). Training and AI support raised it to about 65%.
How many deepfakes are there? Nobody has counted. The common “8 million in 2025” figure is an extrapolation: a 2023 startup estimate of 500,000 deepfakes on social media, doubled every six months.
How much money is lost to deepfake fraud? There is no official total. Deloitte forecasts generative AI could enable $40 billion in US fraud losses by 2027, and single cases have reached $25 million (Arup, 2024).
Are voice deepfakes increasing? Pindrop, which monitors call center fraud, reported a 680% rise in voice deepfakes in 2024 (Pindrop, 2025). The FBI has warned of AI voice messages impersonating US officials.
Related: Cybersecurity Statistics · AI Statistics · Email Statistics · Call Center Statistics · AI in Healthcare Statistics
Ready to automate your busywork?
Carly schedules, researches, and briefs you—so you can focus on what matters.
See what people say
"Before Carly, I relied on a Calendly link, but the whole process felt impersonal and not very professional. Carly changed that by handling all the back-and-forth, so I'm no longer stuck in endless email threads trying to line up schedules.
Now Carly reaches out to candidates, shares my real-time availability, lets them pick a slot, then sends a Zoom link and drops it straight into my calendar. She sends reminders to both of us before each call, which has significantly reduced no-shows and last-minute confusion.
On top of scheduling, Carly acts like a full executive assistant, sending me my schedule the night before so I can prepare for each call. It reminds me of the old x.ai assistant, but Carly is noticeably smarter, faster, and better suited to my healthcare recruitment business."
