AI Shield: A Review of Browser-Based Multi-Threat Detection for AI-Generated Content, Phishing, and Deepfake Images

Ordinary internet users face a growing problem: AI-generated text, deepfake images, and phishingmessages have all become convincing enough that manual identification is no longer a reliable defence.Existing tools address each threat in isolation, often require cloud connectivity, and create enoughworkflow friction that most users simply don't bother. This paper reviews the design, implementation,and evaluation of AI Shield — a Chrome browser extension backed by a locally-running FastAPIserver that combines AI text detection, deepfake image classification, and phishing analysis in oneinterface. The text detection module, built on a fine-tuned DeBERTa-v3 transformer, achieves 91%accuracy on human-authored samples and 89% on GPT-4-generated outputs. The image module, usinga ResNet-50 classifier, reaches 93% accuracy on real photographs and 88% on GAN-generated faces.A hybrid rule-plus-NLP phishing pipeline hits a 91% detection rate on phishing emails with a falsepositive rate of 6%. All inference runs locally; no user content is transmitted to external servers. Thepaper covers the motivating context, related work, system architecture, detection methodology,experimental results, and the main limitations and directions for future development.