AI Web Security Guard

“Real-Time Protection. AI-Powered Confidence.”

Comment

✅ Inspiration

The internet has become an essential part of everyday life, but with it comes a rise in online threats — phishing pages, scam forms, malicious pop-ups, and unsafe URLs. While working on regular browsing tasks, I noticed that most users (especially beginners, students, and elderly people) cannot easily differentiate between a safe website and a harmful one.

This inspired me to build an AI-powered tool that protects users automatically without requiring technical knowledge. The goal was simple:

“Make web safety intelligent, automatic, and effortless.”

Chrome’s new built-in AI capabilities motivated me to explore how on-device AI can be used to analyze webpages in real time.

✅ What It Does

AI Web Security Guard is a Chrome extension that:

Scans webpages instantly using on-device AI

Detects phishing patterns, suspicious URLs, and unsafe content

Assigns a risk score based on AI evaluation

Shows results as:

✅ Safe

⚠️ Moderate Risk

⛔ High Risk

Protects users without sending any data to external servers

Works quietly in the background while browsing

Mathematically, the risk score is computed using a weighted function:

𝑅

𝑤 1 𝐶 text + 𝑤 2 𝐶 links + 𝑤 3 𝐶 forms R=w 1 ​

C text ​

+w 2 ​

C links ​

+w 3 ​

C forms ​

Where:

𝐶 text C text ​

= AI classification of malicious text

𝐶 links C links ​

= suspicious link score

𝐶 forms C forms ​

= fake login form probability

𝑤 𝑖 w i ​

= importance weights

If 𝑅

0.8 R>0.8, the page is flagged as High Risk.

✅ How We Built It

The project was built as a full Chrome Extension using:

Manifest V3

Content scripts to extract visible text and URLs

Background service worker to run AI analysis

Chrome Built-in AI APIs (2025) for classification

Popup UI to display risk results

Local storage to save logs and score history

Technical workflow:

Content script extracts webpage content

Cleaned text and URLs are sent to the background

AI classifier computes risk using on-device models

A risk score is generated

The popup displays real-time results

No external database or server was used — everything runs locally.

✅ Challenges We Ran Into

Understanding how Chrome’s new built-in AI APIs work

Managing cross-file communication between content scripts and background worker

Keeping performance fast even on websites with heavy text

Avoiding false positives while still detecting real threats

Handling asynchronous messaging inside the extension

Ensuring the UI updates instantly based on the AI result

At times the AI model was too sensitive or not sensitive enough, and tuning the weights required trial and error.

✅ Accomplishments That We’re Proud Of

Successfully integrated Chrome’s AI into a real, working extension

Achieved real-time page scanning without slowing down browsing

Created a privacy-focused security tool that runs entirely on-device

Designed a clean, simple UI that anyone can use

Packaged a full Chrome extension ready for deployment

Built a reliable risk-scoring system using both AI and rule-based logic

✅ What We Learned

Throughout the project, we learned:

How modern Chrome extensions are structured with Manifest V3

How on-device AI models can be integrated into browser workflows

Best practices for securing user data

Techniques for text extraction and URL analysis

Asynchronous communication between extension components

The importance of lightweight UI/UX in browser tools

We also gained experience in evaluating web security signals mathematically:

Final Decision

{ Safe ,

𝑅 ≤ 0.5

Moderate ,

0.5 < 𝑅 ≤ 0.8

High Risk ,

𝑅

0.8 Final Decision= ⎩ ⎨ ⎧ ​

Safe, Moderate, High Risk, ​

R≤0.5 0.50.8 ​

✅ What’s Next for AI Web Security Guard

In future versions, we plan to add:

✅ AI-powered detection of fake OTP and payment pages

✅ Voice alerts for visually impaired users

✅ Badge icons that change color based on risk

✅ A dashboard showing browsing safety history

✅ Multi-language threat detection

✅ Chrome AI vision models for scanning page screenshots

✅ Auto-block mode for extremely dangerous links

The goal is to evolve this project into a full real-time AI-powered browser safety companion.

Built With

Share this project:

Updates

Submission history