PhishGuard Senior NeuralCiberGuard · Product 01
Product Preview · Interactive

Phishing protection that doesn't leave grandma behind.

A browser extension built for adults 65+. Three-signal AI detection with accessibility-first design. When something looks wrong, one big button alerts your family — with a screenshot — before any money moves.

$4.8B
Lost to fraud by adults 60+ annually
Higher phishing susceptibility vs. under-40 group
WCAG 2.1
AA compliance — designed in, not bolted on
Standard: W3C WCAG 2.1 AA
Zero
Competitors combining ML + a11y + family alerts
Market survey conducted Q1 2026 · NeuralCiberGuard research

See how it thinks.

Click any scenario below to see PhishGuard evaluate a real-world situation. The extension runs three detection signals in parallel and explains its verdict in language anyone can understand.

https://www.rbcroyalbank.com/sign-in

Three detection signals. One clear verdict.

Most phishing tools rely on blocklists that can't catch new attacks. PhishGuard combines three independent ML signals so novel phishing gets caught on first contact — not after hundreds of victims report it.

Signal 1 — URL Features

Domain age, SSL certificate age, Levenshtein distance from 200+ known brands (catches amaz0n.com, paypa1.com, rbc-login-verify.com), TLD risk scoring, suspicious URL parameters.

XGBoost · python-whois · 50K labeled URLs · cached 7 days

Signal 2 — DOM Analysis

Runs in the page. Detects form actions that post to a different domain, hidden iframes that overlay real sign-in buttons, and Browser-in-the-Browser attacks where the "browser chrome" is just HTML drawn on top of the page.

Chrome Manifest V3 · content script · MutationObserver

Signal 3 — Visual Similarity

Perceptual hash of the rendered page compared against a 200-brand library. A page that looks like Bank of America but loads from boa-secure-login.ru gets flagged immediately — even if the URL features were inconclusive.

imagehash · pHash 64-bit · 200-brand fingerprint DB

Ask My Family — One Button

When in doubt, one press captures a screenshot and sends it via SMS to a pre-configured trusted contact. Works even when the user hasn't been warned — because sometimes the scam slips past ML, but never past a skeptical grandchild.

Twilio Programmable SMS · HMAC-signed screenshot upload

Accessibility First — Not a Retrofit

48px+ shield icon, 18px+ body text, 4.5:1 contrast minimum, keyboard navigation throughout, screen-reader announcements on verdict changes, respects prefers-reduced-motion. Tested with real seniors center users.

WCAG 2.1 AA · axe-core verified · VoiceOver + NVDA tested

Privacy — Yours Stays Yours

URLs are hashed before any ML backend sees them. Screenshots never leave the device except when the user presses Ask My Family. No analytics, no ad SDKs, no telemetry you didn't opt into.

Local-first · SHA-256 URL hashing · Twilio E2E only on request

How it's built

Chrome Manifest V3 + Firefox MV3 compatible. ML inference runs in a Flask backend because MV3 restricts ML libraries in extensions. Everything else runs on-device.

01
Extension (MV3)
  • Background service worker
  • Content script per tab
  • MutationObserver on DOM
  • Popup UI (vanilla JS)
02
ML Backend
  • Flask + Gunicorn
  • XGBoost URL classifier
  • imagehash pHash engine
  • python-whois cached 7d
03
Training Data
  • PhishTank (50K URLs)
  • Tranco top 1M (benign)
  • 200-brand pHash library
  • Elderly-targeted scam corpus
04
Family Alert
  • Twilio Programmable SMS
  • HMAC-signed screenshot URL
  • 24h expiring R2 link
  • Fallback: email via SES
Screenshot sent to Sarah (your daughter)