Detecting Deepfakes: Here’s How Users Can Analyse Fake Content

In a world where trust is vital, deepfakes can spread lies. Imagine seeing a video of a politician saying something outrageous. Many people might believe it, which could lead to chaos or fear. Misinformation spreads like wildfire, and deepfakes can add fuel to this fire.

Deepfakes are affecting individuals and organisations alike. A widespread deepfake video scam is leveraging famous Indian personalities, including Virat Kohli, Mukesh Ambani, Anant Ambani, and Neeraj Chopra, along with international figures, to promote fraudulent mobile gaming applications, according to cybersecurity firm CloudSEK.

When it comes to businesses, over half of C-suite and other executives (51%) expect an increase in the number and size of deepfake financial fraud over the next 12 months, according to a recent survey by Deloitte. 

In the BFSI sector, deepfakes are being used for video KYC fraud, ghost fraud, and stock manipulation. The healthcare sector faces threats like disinformation, manipulated medical records, and doctor impersonation. Government entities are at risk from election interference, incitement of violence, and legal implications due to false evidence.

Deepfakes can further promote fake news and fraudulent advertisements in the news and media sector. They also challenge the IT and telecom sectors, damaging brand reputations and driving financial scams.

In an attempt to identify and mitigate the risks posed by deepfake content, CloudSEK has launched a Deep Fake Detection Technology. The DeepFake Detector calculates an overall Fakeness Score by integrating several analysers.

This includes a video fakeness analyser, which detects manipulated faces in video frames; an audio fakeness analyser, which identifies synthetic sounds and altered speech patterns; Facial Coherence Analyser, which checks unnatural transitions in facial expressions and movements; Texture Uniformity Analyser, which inspects inconsistencies in skin and background textures; and Audio Transcriber and Summary Generator, which transcribes audio and summarises key points.

CloudSEK said that the technology is available for free, and individuals and organisations can check deepfake content through the following process:

• Go to CloudSEK Community.

• Sign in using your GitHub, Google or corporate email account.

• Click on the "Deepfake Analyzer" tool.

• Paste the URL of the video you wish to analyse (Facebook, Instagram, X, YouTube).

• The analyser will assess the video using various modules. This may take up to 2 minutes.

• Check the video's description for key insights, and any organisations or individuals mentioned in the video.

• A Fakeness Score of above 70% implies AI-generated content; 40-70% is Dubious, possibly a mix of original and deep fake elements; and below 40% likely indicates human-generated content.

According to CloudSEK, if the video quality is low (240p or less), the analysis may not be accurate, as the frames may be unrecognisable.