Recently, two images have garnered widespread attention on social media. The first one features Instagram celebrity Zara Patel in a genuine video, while the second showcases a fabricated video using deepfake technology, superimposing Rashmika Mandanna's face. This fake video of Rashmika was so convincing that it deceived thousands of social media users due to its authentic-looking expressions.
Notable figures, including Amitabh Bachchan and Rashmika Mandanna herself, expressed their astonishment. In a Twitter post, Rashmika wrote, "To be honest, something like this is not only terrifying for me but for everyone. If such a thing had happened to me during my school or college years, I would not have known how to cope with it."
For a comprehensive understanding of deepfake technology, its mechanics, and the regulations surrounding it, you can delve into the detailed explanation. But let's start by clarifying what deepfake is and how it is produced.
What is Deepfake?
Deepfake, a term coined in 2017, involves inserting someone else's face, voice, and expressions into a genuine video, photo, or audio recording with such precision that it can be indistinguishable from reality. This process leverages machine learning and artificial intelligence, enabling the creation of realistic videos and audio using technology and software.
AI and cybersecurity expert Puneet Pandey points out that readily available technology and software packages have made deepfake creation accessible to anyone. Furthermore, voice cloning has become a significant concern due to advancements in current technology.
How to make a deepfake video?
Creating a deepfake video involves several steps. First, a suitable image or video is chosen, from which a fake face and voice can be integrated. The encoder algorithm is used to identify and gather the faces, while the decoder algorithm swaps the faces and adds expressions. Additionally, audio is incorporated, and extensive work is done on each frame of the video to make it appear authentic, with a standard video consisting of 24 frames per second.
The advent of AI has made deepfake creation even easier, with various public domain apps allowing users to upload a photo and generate a fake video or photo of the chosen person. However, experts caution that videos produced with flawless AI technology can deceive even the most discerning viewers. By the time the fake video's true nature is revealed, the subject's reputation may be irreparably damaged. Research conducted in the United States indicates that approximately 96% of AI-based deepfake videos are used for creating explicit content.
Those who create fake videos continuously outpace those trying to detect them. For instance, when American researchers reported in 2018 that fake videos did not depict blinking eyelids, the forgers soon began incorporating this detail.
There is currently no certified tool to identify deepfake videos, but several companies, including Google, Amazon, and Meta, are actively working on developing detection methods. Nonetheless, there are ways to uncover the authenticity of a video by paying close attention to certain details.
Law in India related to Deepfake
As of now, there is no specific law in India to address the issue of deepfake content. The European Union has introduced the "Code of Practice on Disinformation" under the AI Act to combat deepfakes, with technology companies like Google, Meta, and Twitter pledging to take action against deepfakes and fake accounts on their platforms after signing this agreement. Companies have six months to implement these measures, and violating the law can result in fines amounting to 6% of their annual global revenue.
In the United States, the Deepfake Task Force Act has been implemented to conduct annual assessments of deepfake technology, but it is not yet rigorously enforced. Additionally, California and Texas have enacted separate laws related to deepfakes.
In India, there is no specific legislation or policy addressing deepfakes, and such cases are typically handled under the IT Act. Under this act, individuals disseminating such videos or information can face imprisonment for three to ten years.
If someone becomes a victim of deepfake content, they can also file a complaint with the local police, the cyber cell, or the website cybercrime.gov.in. Law enforcement agencies can trace IP addresses and locate the individuals behind these activities, potentially leading to their prosecution. Complaints about cybercrimes can also be reported via the toll-free number 1930.
Now, let's explore an incident related to deepfake content and some notable examples of deepfakes that have gone viral.
In a case from Guwahati, Assam, a girl was informed by a friend that her private photos were being sold on a website. She and her father filed a complaint with the police, but the authorities initially dismissed it as a common complaint. The girl then took matters into her own hands and conducted her investigation.
She was shocked to discover that her seemingly private photos, which had been manipulated to appear nude using AI, were not unique to her but also included images of thousands of women from across the country, all of which appeared entirely genuine. When she approached the police again, this time explaining the concept of deepfake technology, the authorities were unfamiliar with the term. The girl subsequently exposed the reality by sharing both her real and manipulated photos from her social media account, shedding light on the presence of a nefarious gang in Assam and raising awareness about the need for caution among young women.
Deepfake technology is primarily used for various purposes, such as creating explicit content, scams, fraud, defaming celebrities, manipulating elections, social engineering, and financial fraud.
Here are three notable examples of deepfake incidents:
1. In March 2022, a fake video emerged in which Russian President Vladimir Putin announced a ceasefire in the Russia-Ukraine war. Similarly, Ukrainian President Volodymyr Zelensky was shown apologizing to Putin. Both videos were later revealed to be deepfakes and were removed from Meta and YouTube.
2. In May 2023, a viral image depicted police arresting protesters at Jantar Mantar, with Vinesh Phogat and Sangeeta Phogat seen smiling. Subsequently, it was exposed as a deepfake image created with AI.
3. In July 2023, a video featuring Seema Haider, who had come to India from Pakistan due to her love for Sachin, was portrayed as a spy. Her real name was Major Samia Rehman, and the video was exposed as a deepfake, with a fake UN photo included in the narrative.
Deepfake is increasing rapidly which needs serious consideration to combat it.