Back to Knowledge Base
DeepShield AI 5 min read

DeepShield AI: Security Model

Technical deep-dive into our hybrid local/cloud AI analysis engine for deepfake detection.

The Hybrid Analysis Engine

Trustity DeepShield utilizes a unique "Edge-First" architecture to detect synthetic media (Deepfakes) and phishing attempts in real-time.

1. Local Inference (The "Fast Path")

A lightweight WASM (WebAssembly) model runs directly inside the user's browser (using Chrome's Offscreen API). It analyzes:

  • Visual Artifacts: Irregularities in pixel distribution common in GAN-generated faces.
  • Text Patterns: NLP heuristics to detect social engineering language (urgency, financial threats).

2. Cloud Intelligence (The "Heavy Lift")

If the local model is uncertain (confidence score < 80%), a perceptual hash of the content is sent to the Trustity Cloud. Our cloud engine compares it against the Global Threat Intelligence database, powered by signals from VirusTotal and our proprietary sensor network.

Last updated: 12/5/2025

Was this helpful?