CIVIC: The Client-side AI Guardian (eRaksha AI for Peace & Safety)

TL;DR: CIVIC is your smart, on-device friend that whispers "Hey, this looks fishy" when you encounter digital arrest scams, fake news, or phishing links—without compromising your privacy. Built for eRaksha AI for Peace & Safety Building Trust in the Age of Machines Initiative

CIVIC (Client-side Intelligent Veracity & Integrity Companion) is like a smart friend living inside your phone and browser whose only job is to whisper, “Hey, this looks fishy, be careful.”

It quietly reads or listens on your device to what websites, messages, and callers are saying, and looks for danger signs: fake news tricks, scam links, bullying words, or classic “digital arrest” scripts where someone on WhatsApp pretends to be police or customs and demands money.

When it sees these patterns, it doesn’t spy on you or upload your chats; instead, it shows a small card on your screen explaining in simple language why this looks risky (for example, “unknown number + threats of arrest + demand for quick payment”) and suggests safe next steps like “hang up and call the official helpline” or “double-check this news on trusted sites.”

Quick: What, How, and Why?

• What: CIVIC, a dual-mode (web + call) client-side AI guardian against misinformation and cyberfraud.
• How: On-device ML, privacy-preserving architecture, overlays in browser and call UI, real-time risk scoring and explanations.
• Why: Directly tackles fake news and “digital arrest” scams hurting citizens, while aligning with government safety goals and privacy regulations.

Key Capabilities

Web Defense

Highlights likely fake news, scam websites, phishing forms, and abusive comments in the browser/AI sidebars.

Call Defense

Warns during suspicious WhatsApp/VoIP/phone calls where someone pretends to be police/customs/CBI and threatens “digital arrest” or fake legal action.

Explanations

Uses simple overlay cards to explain why something looks risky (language patterns, domain issues, scam script patterns) and suggests safe actions.

Why This Matters

India and many countries are facing a surge in deepfake, AI scams and “digital arrest” style fraud. The Supreme Court, NITI Aayog, and banks have issued specific warnings, but a client-side, privacy-first assistant helps citizens directly at the point of risk, without waiting for platforms to moderate everything.

System Architecture

CIVIC operates entirely client-side to preserve user privacy. It acts as a guardian layer between the user and potential digital threats.

User [Citizen] --> Browses Web / Receives Call

[CIVIC Guardian Layer (Local Device)]
Browser/Phone --> On-Device AI Engine
AI Engine --> Risk Score --> Safety Decision Logic

Decision:
-> Safe: Green Verification Badge
-> High Risk: Red Alert / Interruption -> User
                    

Implementation & Future

Phase 1 (Web MVP): A Manifest v3 extension running small transformers (DistilBERT/TinyBERT) via ONNX/WebAssembly to compute misinformation scores locally.

Phase 2 (Mobile MVP): An Android app using accessibility permissions to detect vishing scripts on the fly.

Phase 3 (Privacy): Integrating Federated Learning to improve models without raw data leaving your phone.

This project aligns with the vision that on-device AI is a practical alternative to cloud-centric systems for privacy-sensitive domains.