FactGuard AI: Real-Time Fake News Detection System for Browsers

FactGuard AI is a machine learning-based Chrome browser extension designed for real-time fake news detection. It identifies false information through text classification, multi-source verification, and image forensics analysis, providing explainable AI results to help users instantly distinguish the authenticity of information while browsing web pages. This project is open-source, maintained by shubhvish22, and hosted on GitHub.

In the digital age of information explosion, fake news spreads quickly and widely, and social media algorithms accelerate the diffusion of false information. The traditional "post-hoc debunking" model is lagging and has caused widespread impacts. FactGuard AI emerged as a solution, shifting detection from "post-processing" to "real-time protection". Through the browser extension, users can verify information while browsing, changing the way information is interacted with.

The system uses a client-server architecture (Chrome extension frontend + Python Flask backend). Text verification module: TF-IDF feature extraction + logistic regression classification, combined with Google News RSS multi-source cross-verification and optional Google Fact Check API. Image verification module: Perceptual hash (pHash) to detect duplicate/modified images; AI-generated image detection uses multiple techniques such as noise variance analysis, texture block analysis, and color distribution entropy for comprehensive scoring.

Diverse interaction methods: selected text analysis, right-click menu operations, image upload/URL pasting, page scanning for batch detection. Visual feedback uses glassmorphism design: image badge labels (FAKE/REAL), red/green borders, hover prompts, three risk levels (HIGH/MODERATE/LOW). It provides explainable AI results, stating the basis for judgment (model confidence, missing sources, image anomalies, etc.).

Backend tech stack: Flask framework, Flask-CORS for cross-origin support, scikit-learn (TF-IDF + logistic regression), Pillow/NumPy for image processing, feedparser for RSS parsing, SQLite for storing image hashes. Frontend: Chrome Manifest V3, Service Worker/Content Script/Popup Script, custom CSS for glassmorphism. Deployment automatically downloads datasets to train models; configuration files allow adjustment of parameters such as ML thresholds, image thresholds, and API keys.

Application scenarios: Passive protection for individual users (identifying political news, health rumors, etc.); quick verification of sources and images for content creators/journalists; teaching cases for educational institutions (integrating multiple technical fields). Value: Saves verification time, avoids spreading false information, and facilitates learning of engineering practices.

Current limitations: Mainly supports English; TF-IDF + logistic regression is insufficient for complex semantics; vulnerable to adversarial attacks; relies on the stability of external services. Improvement directions: Multi-language support, upgrade to deep learning models (e.g., BERT), community crowdsourced feedback, offline mode, social media integration.

FactGuard AI provides a practical approach to tackling fake news, encapsulating complex technologies for ordinary users. Its open-source nature avoids monopolization, and transparency and auditability help build trust. With technological evolution and community contributions, it is expected to become a standard configuration for web browsing, similar to how antivirus software is for operating systems.