🎭 Module 3: Deepfakes & AI-Assisted Detection
Welcome to Module 3. As generative AI becomes more accessible, cybercriminals are increasingly using "deepfakes"—synthetic audio, video, and images—to conduct fraud, extortion, and misinformation campaigns.
This module covers the basics of deepfake technology and provides easy, non-technical methods for officers to detect manipulated media.
🛑 Understanding the Threat
Deepfakes rely on deep learning (specifically GANs - Generative Adversarial Networks) to swap faces, clone voices, or generate entirely fabricated events.
- Voice Cloning Fraud: Scammers only need a 3-second audio clip from social media to clone a victim's voice and call their relatives to demand ransom.
- Video Evidence Tampering: Fabricating CCTV footage or political speeches to mislead investigations.
- Identity Spoofing: Creating fake live video feeds to bypass biometric KYC (Know Your Customer) checks at banks.
🔍 Easy Ways to Spot Deepfakes (No Tech Required)
You don't need a degree in computer science to spot a deepfake. The AI often leaves behind "artifacts" (errors) that the human eye can catch if you know where to look.
1️⃣ Facial and Visual Anomalies
- The "Uncanny Valley" Eyes: Deepfakes often struggle with realistic eye movement. Look for lack of blinking or unnatural reflections in the pupils.
- Blurry Edges: Pay attention to the edges of the face, specifically where the hair meets the forehead or the jawline meets the neck. Face-swaps often have a strange, blurred "halo" effect.
- Glasses and Jewelry: AI struggles to render the physics of light reflecting off glasses or dangling earrings consistently frame-by-frame.
2️⃣ Audio Glitches
- Robotic Cadence: Listen for unnatural pauses, lack of breathing sounds, or a slightly metallic "robotic" undertone.
- Emotional Disconnect: The speaker's tone of voice often doesn't match the urgency or emotion of the words being spoken.
🛡️ Non-Technical AI Detection Tools
When the naked eye isn't enough, you can use accessible, free tools to help verify authenticity.
🖼️ Image Verification
- Reverse Image Searching: The simplest way to debunk a deepfake image is to find the original source photo. Use Google Lens or TinEye to see if the base photo exists elsewhere.
- AI Image Detectors: Try AI or Not or ImageWhisperer, free web tools where officers can upload a suspicious image to check if it was generated by AI.
🎥 Video & 🎤 Audio Analysis
- Deepware Scanner: Open-source platform (Deepware.ai) that scans uploaded videos or YouTube links and provides a "Fake Probability" score.
- ScreenApp AI Video Detector: A free online interface where you can drop video links to analyze both face-swaps and lip-sync manipulation.
- Resemble AI Detect: An audio-specific tool designed purely to detect "voice clones." Upload a suspect audio clip to see if the voice is synthetically generated.
📝 Text Checkers
- ZeroGPT: If you suspect a phishing email or a ransom demand was written by AI, paste the text into ZeroGPT. It highlights sentences that exhibit "low perplexity" (a strong indicator of bot authorship).
[!TIP] The Golden Rule: Always verify out-of-band. If a panicked family member calls asking for money from a new number, hang up and call their known original phone number to establish absolute verification.