How deepfakes are used to scam You & Me? Current trends on detection using AI & legal regulations worldwide.

Deepfakes are now impacting common people like you and me. As part of a series on how Artificial Intelligence(AI) is helping cybercriminals, Last week I talked about how cybercriminals are using AI for creating deepfakes for cybercrimes. In continuation, this week I am going to explain with couple of examples how audio-visual deepfakes were used on common people for major cybercrimes outside India(In three different countries) below as shortly it may hit us too in India.

Last week, I gave a overview of how deepfake works and explained some measures you can take to protect yourself and detect deepfakes along with legal(Indian) and other remedies if you were a victim. In this article I am also going to talk about how AI can help in detecting deepfakes and laws or regulations adopted by various countries to regulate deepfakes.

Let’s look at below actual cybercrimes :

• Last month in China, cybercriminals duped INR 5 crores(4.3 million yuan) from the unsuspecting victim by using AI created deepfake video live chat of his friend in need of urgent money for an auction, which he came to know later when his friend denied of any such call.
• In USA, One lady received a call by a kidnapper informing her that her daughter is with them and the lady was able to clearly hear her daughter’s voice in the background screaming for help and asked for a ransom of INR 8 crores(1 million USD). Later it was found that daughter was safe and the daughter’s voice was deepfaked.
• The CEO of an energy firm based in the UK thought he was following his German boss’s urgent orders, when he transferred INR 2.3 crores(220,00 UK pounds)funds to a third-party account. But the request actually came from the AI-assisted deepfake voice recording on phone.

How the above deepfake cybercrime was carried out :-

A good deepfake video with audio can be created on a decent home computer with good Graphic Processing Unit(GPU) card or with a good midrange and above smartphone and a deepfake creation software or online. Several minutes of 4K source and destination footage are required and the target videos should demonstrate similar ranges of facial expressions, eye movements, and head turns. One final important point is that the identities of source and destination should be similar physically.

For just audio deepfake, a 5 second audio clip of target person is good enough, from that software extracts the key characteristics of the voice, as well as the intonations and emotions that make that voice unique.

Victim’s many pictures or videos along with target pre-recorded video are fed into the deepfake software, which then breaks them into frames, and the victims face is swapped on target video actor’s face in each frame. The frame which doesn’t fit or wrongly superimposed is removed, and the software’s algorithm is trained to pick and improve on the expressions framewise and this process is repeated till a good video is created. Then text(script) to audio conversion with victim’s audio is done using the picked up unique characteristics. After that lip syncing is done as per the script and the created audio. Thereby creating the intended deepfake video along with the audio.

How AI can help in detecting deepfakes:

Like the famous proverb – “use a thorn to remove a thorn”, AI can be harnessed to detect deepfakes, and there has been lot of work going on this in recent time and as per INTEL Corp, its deepfake detection software can detect up to 96% of deepfakes out there. There are many technical ways to detect deepfake like:

• The AI tool can analyze a still photo or video detect whether or not the media is manipulated and provide a confidence score. This is what the Microsoft’s Video Authenticator AI tool does.
• The researchers from Binghamton University and INTEL Corp. have created a AI tool, which looks for 32 unique biological and generative noise signals called ‘deepfake heartbeats’ left by deepfake model videos, to determine if its a deepfake.
• The AI tool proposed by Stanford University and the University of California looks at dynamics of the mouth shape movements and lip sync for detecting deepfake.
• Few other AI tools are based on the premise that AI technology which churned out deepfakes, can be altered to separate the real or the imposed image/audio from the source.
• Another regulatory technique, which can be of help is that every deepfake creation tool must post a unique signature on every creation, which can be later used for detection.

Laws and Regulations on deepfake cybercrimes worldwide :-

• India currently does not have any law or regulation on AI or deepfakes directly, but as explained last week there are provisions under IPC and IT Act which can be looked at for legal remedies, But the new proposed Digital India Act will have a chapter on AI and deepfakes in particular as per the Minister of State(MoS) Rajeev Chandrasekhar¹.
• China’s new regulations this year, called Deep Synthesis Provisions, govern deep synthesis (or deepfake) technology and services, including text, images, audio, and video produced using AI-based models². This appears quite comprehensive.
• Several US states have passed laws for banning deepfakes related to political elections or pornography.
• European Commission has proposed the AI Act, which includes deepfakes under its scope in an initial attempt to regulate them. The act has yet to be passed though.

Frequently Asked Questions :

If you find this useful, please like it and share, also if you have any questions or topics you want me to cover please add them in the comments section. Also you can share one page poster covering the entire content with your friends and relatives in other social media.

References :-

1. MoS Rajeev Chandrasekhar holds Digital India Dialogues in Mumbai on the Principles of the Digital India Act
2. China’s New Legislation on Deepfakes.
3. EU AI Act: first regulation on artificial intelligence