Can You Spot a Deepfake?

Deepfakes have taken the world by storm in recent years, with searches for the term emerging in early 2018. We’ve heard the term banded around a lot in the media, and the UK Government has even taken steps to improve the public’s knowledge surrounding deepfake videos. Understanding how deepfakes are created and how deepfake technology can be used in cyber crime can arm you with the knowledge to stay safe.

Understanding Deepfakes and Their Roll in Phishing Attempts

In essence, a deepfake is a manipulation of an image or video, which has cut-and-paste one person’s face onto another person’s body – sometimes adjusting the words that they say and their facial expressions.

Immediately, the question arises – how can you trust what you see? Let’s investigate what’s behind deepfakes. Whilst the concept of image manipulation has been around for the better part of two centuries, essentially as long as exposable film itself, deepfakes are far cry from your average photoshop edit.

The tools for video manipulation have existed for decades, but it wasn’t until the 1990s that we saw real material development into pre-trained generative video. Hollywood had played with the concept(albeit in-plot only) for 1987’s Running Man with Arnold Schwarzenegger, but the advent of the deepfake didn’t truly come until a decade later in 1997, when Video Rewrite used 2D graphics to cut out and replace segments of video using audio analysis to generate completely new visuals. In effect, the program allowed for users to take full sentences, and put them into someone else’s mouth.

Fast-forward more than 25 years and modern deepfakes are now almost exclusively built by artificial intelligence – a subset of which is “deep learning”, the term from which deepfakes derive their name. These complex computer systems churn through gigabytes of image and audio data to semantically understand a piece of visual media and, once given an appropriate prompt, edit that media.

Some prominent examples of this include the academic “Synthesising Barack Obama Project” in which students at the University of Washington developed a neural network to lip-sync videos of the 44th President of the United States. In film and TV, we’ve seen actors de-aged and resurrected, such as Carrie Fisher and Peter Cushing in Star Wars: Rogue One. More recently, the veracity of online media has become more and more difficult – the moral and ethical implications of deepfakes are worrying, and cyber criminals are all too happy to exploit the new techniques available to them.

The Technology Behind Deepfakes

Modern deepfake technology is almost exclusively developed around artificial intelligence – or “AI”. We’ve spoken at length about the dangers and opportunities that AI bring to the table, from intricate phishing scams to more efficient medical analysis.

Ultimately, all AI does is learn.

Artificial Intelligence can only generate output based on what it has learned. Huge datasets are fed into these computer models in order to train them. The most popular Large Language Models (LLMs) follow this model too- the GPT in ChatGPT stands for “Generative Pre-trained Tensor”, a tensor being a container for different types of information.

Machine Learning, Deep Learning and Large Language Models have all been designed to assist users with tasks. AI does this by crunching through vast quantities of data much faster than a human could.

A few examples of advanced image generation models include:

• Midjourney (Midjourney Inc.)
• DALL-E (OpenAI)
• Sora (OpenAI)
• Stable Diffusion (Stability AI)

Signs to Spot a Deepfake

Deepfakes are now synonymous with scams – after all, they’re literally trying to impersonate someone else.

You can usually head off phishing attempts and cyber attacks from the onset – a deepfake could be used in any number of situations, but if the circumstances are out-of-the-ordinary, stop and think twice.

We know that AI is capable of impersonating the human voice, and that it’s exploited for use in vishing scams (a portmanteau of voice-phishing). Vishing calls are increasingly commonplace, so it’s important to stay alert to suspicious text messages, voice messages or unknown phone numbers or phone calls.

Vishing and deepfakes really are very dangerous – in February 2024, a company in Hong Kong was targeted by fraudsters who used a deepfake via video call, resulting in the company transferring-out HK$200,000,000 – more than £20,000,000.

Spotting a deepfake can sometimes be exceedingly easy – an implausible situation or event would scream “fake” to you – such as an astronaut riding a horse on the moon.

Thankfully, there are tools available to us whereby we can check whether an image is real or not. There are many websites out there that can help you find the origin of an image or video, Google Image Search even provides “Reverse Image Search”. Drag-and-Drop an image into reverse image search, and Google will scour its database for images that appear similar or the same.

Get an Expert Opinion

If you’re really not sure – you can always reach out to an expert. At Fusion IT, we’re specialists in keeping our clients safe, and offer specific training on identifying phishing scams and fake websites.

Impact and Consequences of Cyber Crime

What damage can deepfakes wreck on a business or individual?

Whether it’s across social media accounts, email, or even mainstream media – deepfake videos can be scarily believable. With most internet users finding news via online systems and platforms, the need for due diligence can’t be understated.

Manipulation of Public Opinion

A deepfake of a public figure saying one thing or another on social media platforms can be particularly damaging. Whilst Tom Hanks’ deepfake endorsement of a dental practice in the United States is relatively benign – a political figure or world leader being manipulated by threat actors via deepfake could influence voters or threaten democratic processes as a whole.

The emergence of deepfake pornography also threatens the public image of those in the spotlight – with female celebrities often the target of this criminal content.

This undoubtedly erodes trust in those in the public eye – as the population cannot be certain of what is real and what is not.

Legal and Ethical Concerns

Deepfakes and AI raises many issues when it comes to intellectual property and royalties – as evidenced by the long-running SAG-AFTRA strike in which actors and screenwriters downed-tools whilst negotiations over the use of AI in Hollywood dragged on.

Prevention and Defence Against Deepfakes and Cyber Attacks

Sadly, many people have fallen victim to identity theft, having their personal details and sensitive information conned from them by deepfakes.

Staying aware of the different types of synthetic media and other techniques that cyber criminals will use to gain access to your confidential information can set you in good stead to stay safe online.

If you need any advice on identifying Deepfakes, please reach out.

Thanks,

Richard

Interested? Call us on 0333 241 4123 or email [email protected] for professional, impartial advice.