Skip to content

Home

Jump to bottom
nsportsman edited this page Feb 6, 2026 · 5 revisions

Augustus - LLM Vulnerability Scanner

Augustus is a comprehensive Go-based LLM vulnerability scanner designed to test large language models against a wide range of security vulnerabilities and adversarial attacks.

Key Features

  • 210+ Vulnerability Probes - Jailbreaks, prompt injection, adversarial attacks, data extraction across 47 categories
  • 28 LLM Providers / 43 Generator Variants - OpenAI, Anthropic, Azure, AWS Bedrock, Google Vertex, Ollama, and more
  • 93 Detection Strategies - Pattern matching, LLM-as-judge, content safety, specialized detectors across 34 categories
  • 7 Buff Transformations - Encoding, paraphrasing, poetry, low-resource language translation
  • 3 Harness Strategies - Probewise, batch, and agent orchestration
  • Single Go Binary - Fast, portable, no dependencies

Quick Start

# Install
go install github.com/praetorian-inc/augustus/cmd/augustus@latest

# List capabilities
augustus list

# Run a scan
augustus scan openai.OpenAI \
  --probe dan.Dan_11_0 \
  --detector dan.DAN \
  --config '{"model":"gpt-4","api_key":"${OPENAI_API_KEY}"}'

# Run all probes
augustus scan openai.OpenAI --all --output results.jsonl

Wiki Contents

Page Description
Probes 210+ vulnerability probes organized by attack category
Generators 28 LLM provider integrations with 43 generator variants
Detectors 93 detection strategies across 34 categories
Buffs and Harnesses 7 buff transformations and 3 harness strategies
Configuration YAML config reference and environment variables
CLI Reference Complete command documentation
Architecture Project structure and internals

Project Goals

  • Achieve feature parity with garak (NVIDIA's Python-based LLM security scanner, 160+ probes, 35+ generators) while providing Go-native performance and single-binary distribution
  • Complement promptfoo (TypeScript LLM evaluation framework, 119 red team plugins, 36 attack strategies, 80+ providers) with a focused vulnerability scanning approach
  • Implement cutting-edge attack techniques from 2024-2026 security research
  • Provide production-ready security testing for LLM deployments
  • Enable CI/CD pipeline integration for continuous security validation

Links

Maintained by: Praetorian

Clone this wiki locally