What Are Deepfakes, and How Can You Spot Them? (2025)

Learn about the rapid advances in deepfake technology, the potential threat to businesses, and possible solutions.

Alyssa Abrams

Senior Content Manager

Pavel Goldman Kalaydin

Head of AI/ML at Sumsub

In the first half of 2023 alone, Britain lost £580 million ($728 m) to fraud. Of this total, £43.5 million ($55 m) was stolen through impersonations of police or bank employees, with £6.9 million ($8.6 m) lost to impersonations of CEOs. These impersonations were done using deepfakes.

In 2024, a finance worker at a multinational firm paid out $25 million after a video call with the CEO. The alleged CEO turned out to be a deepfake, and the money was transferred to fraudsters.

Deepfakes, hyper-realistic synthetic media created using sophisticated AI algorithms, have captured the imagination of the public while raising concerns about their misuse.

Deepfakes today pose significant threats to businesses, ranging from reputational damage to financial fraud. So, how do businesses best protect themselves? Let’s dive into the implications of deepfakes for businesses and how AI technology can be used to identify and combat this growing threat.

What are deepfakes?

Deepfakes are synthetic media in which a person’s likeness is replaced with someone else’s through advanced artificial intelligence techniques, often used to create convincing but false images or videos.

Deepfake examples

Some examples of deepfakes include:

Manipulating politicians’ speeches and/or likenesses to shape public opinion. There have already been some dangerous precedents set, with deepfakes of Joe Biden and Slovak politician Michal Simecka being used to undermine elections in both countries.
Face-swapping actors in movies. One of the most well-known examples of this is ‘Fast 7’ which ‘featured’ Paul Walker—an actor who passed away in 2013 while the movie was in production. Computer Generated Imagery (CGI) was then combined with deepfake technology to create an ultra-realistic imitation of Paul Walker. Other examples of deepfakes in film include ‘Avatar: The Way of Water’ and the ‘The Mandalorian’ season 2 finale, where de-aging software was used.
AI-generated porn. Recent cases involve deepfake pornographic images of Taylor Swift and Marvel actor Xochitl Gomez, which were spread through the social network X. However, deepfake porn doesn’t just affect celebrities. Any person who shares photos online may fall victim to deepfake porn. Read more in this article:

How do deepfakes work?

Deepfakes work by leveraging deep learning algorithms (face swapping) to create realistic fake images, videos, or audio based on someone’s facial expressions, gestures, and voice patterns. These algorithms analyze and synthesize someone’s appearance to generate new content, seamlessly replicating their natural movements and speech patterns. Post-processing techniques may be applied to further enhance the realism of the deepfake, making it increasingly difficult to distinguish from genuine content.

Deepfake statistics

A study by Home Security Heroes revealed more than 95,000 deepfakes circulating online in 2023, up 550% since 2019. It also revealed that 98% of deepfake videos were pornographic, with 99% using a woman’s likeness.

In 2023, the use of deepfakes surged across the globe, and by 2024, they have turned into widely used and accessible fraud tools. According to Sumsub Identity Fraud Report 2024, there has been a 4x increase in the number of deepfakes detected worldwide from 2023 to 2024, accounting for 7% of all fraud attempts.

The results fraudsters can achieve using AI are becoming indistinguishable from reality. For instance, 70% of people in a worldwide McAfee survey said they weren’t confident they could tell the difference between a cloned voice and a real one.

Why are deepfakes dangerous for businesses?

The proliferation of open-source tools and tutorials has made deepfake creation highly accessible, raising concerns about the potential misuse of this technology. Deepfakes therefore pose several dangers for businesses:

Impersonation fraud or internal fraud. Scammers can use deepfakes to impersonate executives or employees, tricking employees into transferring funds, sharing sensitive information, or carrying out other malicious actions, resulting in financial losses or data breaches. In 2024, WPP’s CEO was targeted by a deepfake scam using a voice clone.
Payment fraud. Deepfake videos are also used as clickbait to drive traffic to malicious websites that harvest card payment details, according to research by Stop Scams UK and consultancy PwC.
Fraud networks. ASee if you can determine whether the picture is a deepfake in our Whack-a-Deepfake minigame: For Fake’s Sake. deepfake network involves an individual or a group that creates a collection of manipulated multimedia content, often using advanced artificial intelligence (AI) techniques. These deceptive media can be used to create convincing yet entirely fabricated videos or audio recordings, as well as fake documents, for various fraudulent purposes.
Bypassing verification. Deepfakes can be used to illegally bypass verification for online services. For instance, someone could use a deepfake to impersonate another individual during a video verification process, thereby gaining access to secure systems or sensitive information. Similarly, teenagers can use deepfakes to make themselves look older, thereby bypassing age verification checks on age-restricted sites.
Manipulation of information. Deepfakes can be used to spread false information about products, services, or financial performance, undermining market confidence or affecting stock prices.
Brand misuse. Deepfakes can be used to create counterfeit advertisements or promotional materials, diluting brand integrity and misleading consumers.
Reputational harm. Deepfakes can be used to create convincing videos or audio recordings of business leaders saying or doing things they never actually said or did, leading to reputational harm and damaging trust with customers, partners, and stakeholders.

Why are deepfakes a concern for everyone?

Deepfakes can manipulate public opinion, spread misinformation, and undermine trust in various aspects of society. They can be used to manipulate elections, damage reputations, spark conflict, and lure people into relationships with the intent to deceive.

Just recently, a woman in Scotland was tricked into sending £17,000 ($21,200) to a person with whom she thought she was in an online relationship. Despite being skeptical initially, she was convinced by the videos the scammer sent the victim. These videos appeared to be fake.

Furthermore, deepfakes raise ethical and privacy concerns, as they can violate individuals’ rights and dignity by appropriating their images without consent, potentially causing psychological trauma.

As such, deepfakes have broad implications that affect society and everyone’s well-being, safety, and trust in the digital age.

How to spot a deepfake and avoid content misuse

While there is no way to completely eliminate the risk of deepfake abuse, there are preventative measures both businesses and users can take:

If you suddenly receive a call or a video message from a relative or boss asking for money or sensitive information, double check it with them using other messengers
Implement a tool for additional approval for large transfers. If a large transaction is detected, the tool will request approval from two or three other colleagues
Examine images thoroughly. When spotting deepfakes, the devil is in the details, and you’ll often be able to recognize one by examining the direction of light and shadows on the face, inconsistencies in facial features, blurriness or distortions around the edges of the face, unnatural backgrounds, etc. You can also check the photo’s metadata to see if it has been manipulated or altered. In a video, deepfakes may not include natural blinking or breathing movements
Be cautious about sharing personal photos online. You can limit the number of personal photos you share on social media, or adjust privacy settings to restrict who can access your photos
Avoid sharing high-resolution, unedited photos that could be easily manipulated
Thoroughly read the terms and conditions of all AI apps, ensuring they cannot use your photos after your use of the app itself
Contact local law enforcement agencies when your photos or videos have been stolen and/or used in an inappropriate manner, or if you fell victim to deepfake-related fraud
Utilize smart technology. While AI may pose danger, it’s also a remedy. AI can be used to detect certain visual or audio artifacts in deepfakes that are absent in authentic media (see which ones below).

See if you can determine whether the picture is a deepfake in our Whack-a-Deepfake minigame: For Fake’s Sake.

Defending against deepfakes

The best ways for businesses to spot deepfakes are:

Machine Learning and AI. ML & AI algorithms, such as Sumsub’s Liveness Detection, can outperform humans in spotting enhanced photos. Moreover, in October 2023, Sumsub released “For Fake’s Sake”, a set of machine learning-driven models that enable the detection of deepfakes and synthetic fraud. This tool is available for free to download and use by all.
Behavioral Analytics. This process monitors unusual patterns of behavior, such as multiple accounts opened with the same Social Security Number or inconsistencies in identity information.
Fraud Networks Detection. You can uncover interconnected patterns of suspicious activity on your platform using Sumsub’s AI-powered Fraud Network Detection solution. This tool provides you with the ability to identify fraud networks before the onboarding stage through AI, allowing you to apprehend an entire fraudulent network rather than just a single fraudster.
Multi-layered defense. To successfully combat deepfake scams, businesses should employ multiple layers of protection, including AI-powered analysis, biometric authentication, and real-time behavior monitoring. These measures help secure the entire user lifecycle—from onboarding to transactions—ensuring robust defense against evolving threats.