Scam detection + AI agent firewall
Check the message before you trust it. Keep unsafe instructions away from GhostGPT.
GhostFilter protects people from scams and protects AI agents from prompt injection, jailbreaks, secret extraction, and unsafe tool-use instructions.
Example result
A suspicious account-verification message
Scam likelihood
92%
Message excerpt
Your account has been suspended. Verify your password immediately using the link below.
Recommended action
Do not use the link. Open the official service directly and check your account there.
Two protection layers
One scanner for humans. One firewall for AI agents.
Scam detection catches social-engineering risk. GhostGPT protection catches prompt injection before untrusted content enters an AI agent context.
Scam detection for people
Paste suspicious messages, emails, links, screenshots, or PDFs and get a clear verdict powered by classifier, rules, threat intel, and AI review.
Prompt-injection firewall for GhostGPT
Detect instruction overrides, jailbreaks, secret-extraction attempts, and unsafe tool-use requests before content reaches an AI agent.
Safe context handoff
Generate a sanitized wrapper so GhostGPT treats external content as untrusted data, not instructions.
Ghosti safety assistant
Ghosti explains each result in normal language, suggests the next safe move, and helps turn risky content into safer GhostGPT context.
AI/ML safety engine
More than one model. A layered decision system.
The scanner shows which checks found evidence, so you can see why it reached the verdict.
ML ensemble
A local classifier is combined with behavioral scam signals like payment intent, impersonation, urgency, and secret-code requests.
Agent firewall
GhostGPT content gets pass, isolate, or block decisions plus a safe context wrapper for untrusted data.
Deep review
High-risk items escalate to structured AI review, while link reputation and email forensics add external evidence.
Published developer SDK
Add GhostFilter protection to any app or AI agent with npm.
The ghostfilter-ai package is live on npm with local-first scam detection, GhostGPT prompt-injection checks, safe context wrappers, and a CLI.
import { ghostfilter } from "ghostfilter-ai";
const result = await ghostfilter.protect({
input: untrustedContent,
mode: "full",
});
if (result.verdict !== "safe") {
console.log(result.reasons);
}Connections and scan lanes
Built like a security command center, simple enough for anyone.
Live connectors, upload scanning, and manual paste lanes make the product feel bigger without pretending browsers can read private chats automatically.
Gmail
Available in the scanner.
Drive
Available in the scanner.
GitHub
Available in the scanner.
Outlook
Roadmap integration slot.
Slack
Available in the scanner.
SMS
Supported through manual paste.
Supported through manual paste.
PDF / Image
Available in the scanner.
Designed to look, not touch.
Read only
Connected accounts cannot be used to send, edit, or delete messages.
Server-side tokens
OAuth credentials are never exposed to the browser.
Your decision
Results explain the evidence instead of pretending to be infallible.
Frequently asked questions
Clear answers before you check anything.
GhostFilter is designed to help you decide safely, without pretending automated analysis is perfect.
What can I check with GhostFilter?
You can paste a text message, email, direct message, link, or AI-agent context. You can also upload screenshots, PDFs, text files, and saved .eml emails.
How does GhostFilter protect GhostGPT?
GhostFilter scans untrusted content before it reaches GhostGPT. If it finds prompt injection, secret extraction, jailbreak, or tool-abuse attempts, it recommends pass, isolate, or block, saves the firewall run, and can generate a safe context wrapper.
Who is Ghosti?
Ghosti is the built-in safety assistant. It explains why something was flagged, gives the safest next step, helps draft a safer verification reply, and explains how to pass content to GhostGPT safely.
Where is the AI/ML part?
Scam Shield uses a trained local classifier, deterministic social-engineering signals, link intelligence, email forensics, and selective AI review. GhostGPT Firewall adds a separate prompt-injection and tool-abuse detector.
Does GhostFilter open suspicious links?
No. Links are inspected from the server using redirect checks and external threat-intelligence services, so you do not need to open them in your own browser.
Can GhostFilter guarantee that a message is safe?
No automated checker can guarantee that. GhostFilter explains the evidence it found and gives you a safer next step, but you should still verify unexpected requests through an official channel.
Do I need to create an account?
No. You can use the scanner immediately. Scan history is tied to this browser unless a connected source is used.
What happens when I connect Gmail, Drive, or GitHub?
GhostFilter receives read-only access for scanning. It cannot send messages, edit files, delete content, or act on your behalf.
What should I do after a high-risk result?
Do not click, reply, pay, or share a code. Contact the person or organization through a separate method you already trust, such as its official app, website, or phone number.
Have a message that feels off?
Check it before you click, reply, pay, or share a code.