Every few years, we see a revolutionary technology completely transforming our lives. This time it is artificial intelligence, though AI has existed since the 1950s. AI caught the world’s attention when Google and IBM invented AI-enabled machines that could beat humans at board games. Soon after, we experienced AI on our phones and devices. Remember Siri, Alexa and Google Maps? Then came ChatGPT. While AI is a revolutionary technology, the bad guys (adversaries) are also learning to use it against you, to hack your phones or trick you into parting with your hard-earned money. Have you heard about Deepfake?
Google’s AlphaGo beat the world Go champion, Lee Sedol, in a five-game match between March 9 and 15, 2016. Go is a two-player abstract strategy board game. AlphaGo won four out of the five games. This match was considered a significant milestone in artificial intelligence, as it was the first time a computer program had beaten a world-class Go player without a handicap.
Before that, IBM’s DeepBlue learned millions of chessboard moves to beat the world chess champion, Garry Kasparov, in 1997. In 2011, IBM Watson beat humans at the popular TV game show Jeopardy.
All these machines use some form of AI and machine learning, where a machine learns cognitively and remembers what it learns.
However, world interest in AI peaked with the launch of OpenAI’s ChatGPT tool to the public in November 2022. Within just five days, a million users were hooked on ChatGPT. This AI tool comes under generative AI (gen AI) because it’s trained to respond to human input and generate output. Users ask ChatGPT questions through prompts, written in plain English. The world was fascinated by the human-like responses that ChatGPT generated.
DeepSeek R1 is another GenAI model that caught the world’s attention because of the accuracy and details in its output. Today, there are more than a dozen AI content generators available for you to try and the popular ones are: Anthropic’s Claude, Google Gemini, Microsoft Copilot, Preplexity AI, Jasper AI, Deepseek R1, Meta Llama, Zapier, Cohere, and IBM Granite. OpenAI has also released other models that are specific to certain tasks, with varying degrees of accuracy: o3 and o4 mini. These are also called AI assistants, AI companions, foundation models, and large language models (LLMs).
While genAI models like ChatGPT began by generating textual output, the technology has evolved quickly in the past two years. Now, certain models can generate audio, video and images.
For instance, OpenAI’s Sora model is a creative tool for creating images and videos. This is done through prompts – just describe the image or video you want to create, and voila!
Before Sora we had DALL-E 3, a popular AI image generator known for its ability to create high-quality, realistic images from text descriptions (prompts).
After its multi-billion-dollar investment in OpenAI, Microsoft started to develop its own AI image generator called Bing Image Creator, and later its own version of ChatGPT, called Microsoft Copilot. In fact, Copilot is being heavily marketed by Microsoft and it is now integrated with Windows 11 and Microsoft 365 (the new name for Microsoft Office Online). Copilot has free and paid versions.
Taking the technology to the next level, we saw the emergence of AI speech synthesis tools, also known as text-to-speech (TTS) tools, that use artificial intelligence to convert written text into human-like speech. These tools offer a range of options for generating synthetic voices, with some even allowing for customization and cloning of existing voices. Popular examples are Murf, Synthesia, ElevenLabs, and Speechify. With these tools one can generate life-like cloned human voices.
Speech-to-text AI tools like Otter.AI, TurboScribe, and Revel are useful for transcribing recorded audio clips to text, saving journalists, reporters, and (medical) transcription professionals hours on the gruelling task of manual transcription.
AI video synthesis tools use artificial intelligence technologies to enable users to create videos from various inputs like text, images, and scripts. These tools can generate new video sequences, edit existing videos, and even make videos with AI-generated avatars or voiceovers. Examples are Synthesia (digital avatars), Runway, Vyond (animation), Sora, Luma and Luma. There are other examples, but they are AI-enabled video editors. A popular example is Adobe Firefly.
These tools save production studios a lot of money as they no longer have to hire human actors or voice-over specialists. They could use voice cloning or video synthesis tools like the ones described above.
However, the bad guys learnt to use these tools as well, and then started using them for AI-related crimes such as deepfake audios, photos and videos.
Also see: Davos 2025: Global Leaders Call for Action on Unchecked AI
The Dark Side of AI
Deepfake is a term that often comes up in articles about AI misuse. Deepfake is a photo, audio, or video of a person whose face or body has been digitally altered to appear to be someone else, typically used maliciously or to spread false information.
The term Deepfake is a combination of two terms: deep learning and fake. Deep learning is a type of machine learning based on artificial neural networks in which multiple processing layers are used to extract progressively higher-level features from data. The technology provides detailed output and it can be misused for cloning voices and video to target institutions or individuals for financial gain.
For instance, there is a demonstration deepfake video of President Obama using voice cloning and video synthesis technology. This video was posted on YouTube, and you can view it here: https://tinyurl.com/mst5umf5
In other words, they put words of an impersonator into Obama’s mouth using an AI tool, to create a fake and synthetic video of the former U.S. president.
While the fake video was created to demonstrate the capabilities of the AI technology, the video looks so realistic that an uninformed person is likely to believe that this is Obama speaking.
The bad guys or adversaries also watch these ‘technology demonstrator’ videos, and use them in devious ways to commit cyber fraud against organizations and individuals. Let’s examine some notable examples.
Deepfake Examples
Financial Fraud: A UK-based energy firm fell victim to a deepfake voice scam in 2019. Criminals used AI to clone the voice of the firm’s CEO, convincing an employee to transfer €220,000 to a fraudulent account.
Video Call Scams: In a recent case, scammers used face-swapping technology during a video call to impersonate a trusted individual. This led to a businessman in China losing $622,000.
Video Call Scams are also being used to target individuals into parting with their money and there have been revenge porn and social media incidents.
Revenge Porn and Social Media Manipulation: Deepfakes have been weaponized to create fake explicit content or social media posts, leading to reputational damage and legal consequences. By superimposing images or videos on pornographic content, individuals have been blackmailed to pay money of face the consequences of these doctored videos being posted on their social media feeds and sent to all their family members and friends in their phone address books.
Corporate Espionage: Deepfake technology has been used to infiltrate video conferences, impersonating executives to extract sensitive information or authorize fraudulent transactions.
CNN reported an incident where fraudsters used deepfake technology to impersonate a company’s CFO in a video conference call, resulting in a $25 million loss. The fake CFO avatar, created through AI, initiated a funds transfer request, tricking a finance worker into making the payment. This incident highlights the increasing sophistication of deepfake scams and the need for enhanced verification processes in financial transactions. You can find the story here: https://tinyurl.com/4hf24u3j
Fake Crime Scene Evidence: Deepfakes have been used to fabricate crime scene photos or surveillance footage, potentially misleading investigations and eroding trust in digital evidence.
These are actual examples and they highlight the urgent need for robust detection tools and awareness to combat the misuse of deepfake technology.
Audio cloning: Scammers use AI voice cloning technology to impersonate trusted individuals like family members or financial institutions to trick victims into revealing personal information or sending money. This technology allows fraudsters to create realistic-sounding voice recordings from just a few seconds of audio, making it difficult to discern authentic from cloned voices.
There are verification tools and websites that ascertain whether images, audio and videos are fake or genuine. A popular tool is Sensity, which has 98% accuracy, having detected over 35,000 malicious deepfakes in the last year alone. It also offers educational resources to help employees and investigators recognize deepfake threats, making it a one-stop solution for businesses prioritizing digital security. You will find a list of deepfake detection tools here: https://tinyurl.com/yp9a4hur
Conclusion
You’ve heard of the maxim Seeking is Believing. But in this age of AI and deepfake technology, even doubting Thomases could be easily fooled, and that maxim no longer holds true.
Be suspicious and curious if you receive a voice call, email, or video message from a known friend, family member, or business associate asking you for money. Always call them and verify – ask if they made that call or sent that email to you.
Lastly, use a secret family code or phrase known only to your family members. So the next time you receive a call from someone informing you that your son or daughter has been kidnapped, ask them to give you the secret family codeword. And they wouldn’t be able to do so, because your kin is probably safe and sound. Alternatively, you could hang up and then call your son and daughter to ascertain if they are safe.