attack-selection-guide

What problem does it solve?

Security teams conducting AI red teaming assessments often struggle to select the most effective adversarial attacks for their specific target type, compute budget, and defensive controls, leading to wasted effort and incomplete test coverage.

Core Features & Use Cases

• Target-Type Guided Selection: Provides clear branching for LLMs, agentic systems, and custom API endpoints to match attack methods to the target architecture.
• Budget & Defense Aligned Recommendations: Maps attack algorithms and transforms to available compute budgets and known defensive controls like content moderation or input sanitization.
• Comprehensive Threat Coverage: Includes pairing recommendations for over 20 threat categories including jailbreaks, MCP attacks, RAG poisoning, and exfiltration testing.
• Use Case: A red teamer testing a customer service chatbot with strong rate limiting can quickly identify that the pair attack algorithm paired with base64 encoding transforms is the most efficient approach for their constraints.

Quick Start

Use the attack-selection-guide skill to identify the optimal attack algorithms and transforms for your AI red teaming assessment of a custom LLM chatbot with strong content moderation defenses.