How many deepfakes are created every day?

What happens when someone makes a deepfake of you

Free apps capable of generating convincing face-swap videos are available to anyone with a smartphone. Creating one takes minutes. The technical barrier is gone.

Security Hero's 2023 report found that 98% of all deepfake videos are non-consensual intimate imagery targeting women. Victims have no reliable mechanism to remove this content. It appears on platforms that operate outside legal jurisdiction, and once published it spreads in ways that are effectively impossible to reverse. As of 2024, only a handful of countries had enacted specific criminal laws on deepfake content.

Fraud is the second front. In early 2024, a finance worker in Hong Kong joined a video call with what he believed was his CFO and wired $25 million to criminals. It was a deepfake. Most countries still have no specific deepfake law; the first cases are only now reaching courts.

The hardest facts: 98%, non-consensual, no path to removal

From academic curiosity to a crisis in five years

1. 2014Generative Adversarial Networks (GANs) invented by Ian Goodfellow, the foundational technology
2. 2017First public deepfakes appear on Reddit
3. 2018Sensity AI begins systematic monitoring; ~8,000 deepfake videos catalogued online
4. 201914,678 deepfake videos detected; congressional hearings on deepfake disinformation held in the US
5. 2020Sensity counts 85,047 deepfake videos by year-end; numbers doubling every 6 months
6. 2022Diffusion models (Stable Diffusion, DALL-E) further lower cost of realistic fake media
7. 2023Deepfake incidents grow 10× in a single year (Sumsub); iProov: face-swap injection attacks +704% within 2023
8. 2024Deepfake video call used to steal $25.6M in Hong Kong; over 100,000 deepfakes/month created globally
9. 2025~8 million deepfakes projected shared online (UK government, 16x vs 2023); $1.28B in documented fraud losses (Resemble AI); Resemble AI documents 1,567 verified incidents

Deepfake volume growth: doubling every year since 2018

Deepfake creation grew from 8,000 detected videos in 2018 to over 100,000 per month in 2024, as generation tools moved from requiring powerful servers to running on smartphones in minutes.

Deepfakes within the wider AI-generated media explosion

Deepfakes are a part of the broader AI-generated media wave, but most AI images are benign. Deepfakes are distinguished by the consent problem: the large majority exist to harm specific real people. Both numbers are growing as generation tools get cheaper.

The deepfake threat: technology, consent, and democracy

Detection and harm research: what studies measure

In perspective

Frequently asked questions

How many deepfake videos are created per day?

The modelled creation rate is approximately 380 per hour, or around 9,100 per day. AnythingCounter extrapolates it from the UK government's projection of ~8 million deepfakes shared online in 2025 (up from an estimated 500,000 in 2023) and it covers only publicly indexed content. Private creation via smartphone apps is not captured, so the real number is likely significantly higher.

What percentage of deepfakes are non-consensual?

96-98% of deepfake videos circulating online are non-consensual intimate imagery (NCII) targeting women without consent, per multiple independent reports. Security Hero's 2023 study, which analysed 95,820 videos from 85 channels and 100+ websites, put the figure at 98%. The share has been consistent across all years of systematic tracking.

How has the deepfake landscape changed since 2018?

From late 2018 to late 2020, Sensity AI found 85,047 deepfake videos online, doubling every six months. By 2023, incidents had grown 10x in a year (Sumsub). Creation tools got cheaper and easier: what needed a server and months in 2018 now runs on a phone in minutes.

Which sectors are most affected by deepfake fraud?

Sumsub's 2023 data showed cryptocurrency and fintech accounted for the large majority of deepfake fraud cases. Deepfake-based attacks include identity verification bypass (using a synthetic face to pass a KYC check), celebrity-likeness investment scams, and direct corporate fraud: a 2024 case in Hong Kong saw $25 million wired after a deepfake video conference impersonating a CFO.

Can deepfakes be detected?

Current AI detection tools achieve 80-90% accuracy in lab conditions, but real-world performance is lower. People spot video deepfakes correctly only about 57% of the time, barely better than random guessing. Detection technology is in an arms race with generation technology.