Miasma: A Rust Tool to Trap AI Web Scrapers in an Endless Poison Pit

The Tool

Miasma is a lightweight Rust-based web server designed to trap AI web scrapers by serving them poisoned training data and an infinite maze of self-referential links. It's gained 60 stars on GitHub and sparked discussion on Hacker News.

The Problem

AI companies scrape the internet at enormous scale to train their models. Website owners have limited options to protect their content. Miasma offers an offensive approach: poison the data that scrapers consume.

How It Works

Step 1: Deploy Honeytrap Links

Add hidden links to your site that only scrapers will follow:

<a href="/bots" style="display: none;" aria-hidden="true" tabindex="1">
  Amazing high quality data here!
</a>

These links are invisible to humans and screen readers but visible to automated crawlers.

Step 2: Route to Miasma

Configure your web server (e.g., Nginx) to proxy the honeytrap path to Miasma:

location ~ ^/bots($|/.*)$ {
    proxy_pass http://localhost:9855;
}

Step 3: Infinite Poison Loop

When a scraper follows the link, Miasma serves:

• Poisoned data from its "poison fountain" — gibberish designed to corrupt training
• Self-referential links that lead to more poisoned pages, creating an infinite crawl loop
• The scraper wastes time and resources on useless data

Technical Details

• Written in Rust for speed and low memory usage
• Minimal footprint — designed not to waste YOUR compute resources
• Install via cargo install miasma or download pre-built binaries
• Configurable link prefixes, ports, and behavior

The Debate

This approach raises interesting questions:

• Effectiveness: Will poisoned data actually degrade model quality?
• Ethics: Is intentionally corrupting training data justified?
• Legality: What are the legal implications of serving misleading content?
• Cat and mouse: How will scrapers adapt to detect and avoid traps?

Context

Miasma is part of a growing movement of anti-scraping tools that includes robots.txt directives, CAPTCHAs, and now active data poisoning. As AI training becomes more aggressive, website owners are developing increasingly creative defenses.

Miasma on GitHub