§ Docs

Attack Atlas

The public, executable knowledge base of AI-app attacks. 50+ patterns across 6 categories.

The Attack Atlas is two things in one product: a public SEO asset that ranks for AI-app security queries, and an executable library the scanner uses.

Categories

  • Prompt injection
  • RAG poisoning
  • Tool abuse
  • Agent hijacking
  • MCP exploitation
  • Multi-turn jailbreaks

Each category links to documented patterns with severity, OWASP LLM mapping, real-world impact, and defenses.

Run a pattern against your app

Every pattern page has a Run this against my app button. Picking a verified target starts a scoped Brektra scan that runs only that pattern. The result lands in your dashboard with a replay link, just like a normal scan.

This is faster than running a full scan when you want a specific finding type for a customer demo, a compliance check, or a regression test of a defense.

Counts toward your scan allowance

A scoped Atlas scan is one scan against your plan's allowance, like any other.

CI/CD

You can run a single Atlas pattern from brektra-cli:

brektra atlas direct-instruction-override --target https://app.example.com

This is useful in CI to gate on a specific pattern without paying for a full surface scan.