Building Trust with Federated Learning in a World of Digital Illusions

DR. ARINDAM SARKAR
HOD & ASSISTANT PROFESSOR, DEPARTMENT OF COMPUTER SCIENCE AND ELECTRONICS, 
RAMAKRISHNA MISSION

Deepfake and synthetic fraud have become the new normal now and it needs immedicate attention.

At its core, synthetic fraud involves creating fake identities by blending real and fabricated personal information. Fraudsters frequently use authentic data—such as Aadhaar or PAN details, addresses, or birthdates—and combine them with fictitious names to forge identities that appear legitimate. These identities are then used to pass verification checks, build financial credibility, and exploit systems—all without a real-world victim, making detection far more difficult.

Compounding this threat is the rapid evolution of deepfake technology. Leveraging advanced AI, attackers can now generate hyper-realistic fake images, voices, and videos that mimic real individuals with uncanny accuracy. While ethical AI platforms like ChatGPT are designed to block malicious content, underground tools such as FraudGPT and DarkBart openly assist bad actors—offering tutorials on generating deepfakes, embedding malware, and manipulating videos with flawless lip-sync.

To counter this, advanced deepfake detection systems like Faceoff employ multi- layered AI analysis to monitor both surface web and dark web activity. These systems assess lip movement, audio tone, psychological likeness, and visual anomalies. The most cutting-edge approaches now include behavioral biometrics—tracking facial posture, eye motion, voice inflection, and even subtle physiological signals like heart rate and oxygen levels extracted from video.

However, security must not compromise privacy. Uploading sensitive data for analysis introduces its own risks. This is where Privacy-Enhancing Technologies (PETs) come in— enabling secure analysis through techniques like federated learning, secure multi-party computation, and differential privacy. These allow encrypted features to be examined without exposing raw personal data.

Looking ahead, generative AI can become a powerful ally in fighting identity fraud—but only if paired with adaptive, feedback-driven systems that evolve with emerging threats. In a world where fake faces can cause real harm, the future of digital trust depends on innovation, vigilance, and designing security directly into our technological foundations.