Coming soon — join the waitlist

Stop sensitive data
before it reaches the AI.

A Chrome extension that scans every prompt you type into Claude, ChatGPT, and Gemini in real time. The moment it spots something sensitive, it stops you — locally, privately, with zero data leaving your machine.

You're on the list. We'll notify you when GuardianAI launches.

No spam. Just a launch notification. Unsubscribe any time.

🔒 100% Local ⚡ Real-time detection ✓ No cloud. No login. No prompt storage. 🇿🇦 POPIA-aware Always free
↓ Download extension (beta)
How to install
  1. Extract the ZIP anywhere on your computer
  2. Open Chrome and go to chrome://extensions
  3. Enable Developer Mode (toggle, top right)
  4. Click Load unpacked → select the extracted folder
Inside the extension

Built for the moment
before you press Enter.

GuardianAI Settings panel — compliance mode, block threshold, auto-sanitize
Settings — compliance mode & thresholds
GuardianAI Keywords panel — organisation keyword list
Keywords — your organisation's sensitive terms
GuardianAI Audit Log — last 10 events with risk levels and actions
Audit Log — local event history, exportable as CSV
The origin
"GuardianAI bridges the gap between AI governance requirements and day-to-day AI usage."

As organizations accelerate AI adoption, balancing innovation with security and compliance has become a critical business challenge. Employees increasingly rely on AI tools to improve productivity, automate workflows, and enhance decision-making, creating new pathways for sensitive information to leave the organization.

Traditional approaches to AI governance, including policies, awareness programs, and training provide important foundations but often lack visibility and control at the point where data is shared. This creates a gap between governance requirements and day-to-day AI usage.

GuardianAI bridges that gap by delivering real-time protection directly within the browser. The platform continuously monitors interactions with AI applications, identifies sensitive information before it is submitted, and helps organizations enforce governance policies without disrupting productivity.

By combining proactive data protection with seamless user experiences, GuardianAI enables enterprises to embrace AI confidently while reducing compliance risk, safeguarding intellectual property, and maintaining control over sensitive business information.

The result is practical, scalable AI governance that empowers organizations to unlock the benefits of AI while meeting security, privacy, and regulatory requirements.

How it works

Five steps.
Zero data leaves your browser.

1
You type a prompt
GuardianAI watches every keystroke and paste event in real time inside Claude, ChatGPT, and Gemini.
2
Local scan fires
After a brief pause, the detection engine runs — 9 pattern types plus your custom keywords — entirely inside your browser.
3
Risk assigned
HIGH blocks submission instantly. MEDIUM shows a warning modal. LOW is completely silent. NONE — you never know it ran.
4
You decide
Replace sensitive values in place, send anyway, or ignore and finish typing before scanning again.
5
Event logged locally
Timestamp, site, risk level, and action stored in your browser only. Export as CSV. No prompt text ever recorded.
Detection

What GuardianAI
catches before you send.

SA ID Numbers
8001015009087
HIGH · §26 POPIA
AWS Access Keys
AKIA···EXAMPLE
HIGH · §19 POPIA
JWT Tokens
eyJhbGci···
HIGH · §19 POPIA
Private Key Blocks
-----BEGIN RSA···
HIGH · §19 POPIA
Connection Strings
mongodb://user:pass@···
HIGH · §19 POPIA
Email Addresses
jon@company.co.za
MEDIUM · §11 POPIA
SA Phone Numbers
+27 82 123 4567
MEDIUM · §11 POPIA
Custom Keywords
CONFIDENTIAL, Project X
MEDIUM · configurable
Source Code
function getUser() {···}
LOW · silent
Warning UI

Three levels.
Always your choice.

HIGH
Blocked
Submission is stopped. A red banner appears. You cannot send until you dismiss it, review the input, or replace the sensitive values in place.
MEDIUM
Warning modal
A modal shows you exactly what was detected, the actual values found, and a plain-English explanation of why it was flagged. You decide what happens next.
LOW
Silent
Source code patterns are flagged internally but cause no interruption. You can send normally.
🚨
High-risk content detected — submission blocked
SA ID number · 1 instance · §26 POPIA
Dismiss & Review
⚠️ Sensitive content detected
Email address · 3 instances · §11 Personal Information
jon@work.co.za sarah@company.com +1 more
💡 Email addresses are personal information that could identify an individual under POPIA §11.
✏️ Replace All Dismiss & Review Send Anyway Ignore for now
POPIA compliance

Built for
South African users.

Most data governance tools are built for enterprise budgets and foreign regulations. GuardianAI is different — it has POPIA baked in from the ground up, designed for South African individuals and teams who are navigating AI adoption carefully.


When POPIA mode is on, every detection is tagged with the relevant section of the Act, warning language shifts to reflect your compliance obligations, and "Send Anyway" is disabled on HIGH risk events — consistent with the duty of care required by the legislation.

POPIA mode helps identify potentially regulated data. It does not guarantee legal compliance. Consult your Information Officer for guidance specific to your organisation.
§11
Conditions for Lawful Processing
Email addresses and phone numbers. Personal information may only be processed with consent or for a legitimate purpose.
§19
Security Safeguards
API keys, tokens, connection strings, private keys. Responsible parties must secure the integrity and confidentiality of personal information.
§26
Special Personal Information
SA ID numbers. Processing is prohibited unless specific conditions under the Act are met.
Privacy by design

No cloud.
No exceptions.

🔒
100% Local
All scanning happens inside your browser. Zero data leaves your machine at any point.
🚫
No Cloud
No server. No API calls. No telemetry. Verified via Chrome DevTools Network tab — zero outbound requests from the extension.
📋
No Prompt Storage
The audit log stores metadata only — timestamp, site, risk level, action. Never the content of your prompts.
⚙️
Fully Configurable
Custom keywords, detection thresholds, auto-sanitize behaviour. Your settings stay on your machine.
Waitlist

Be the first
to know.

GuardianAI is in private pilot. Join the waitlist and we'll notify you the moment it's available on the Chrome Store.

You're on the list. We'll notify you when GuardianAI launches.

No spam. Just a launch notification. Unsubscribe any time.