What problem does it solve? Applications increasingly embed AI features like chatbots, RAG knowledge bases, and agent tool calling, but traditional web vulnerability testing does not cover their unique attack surfaces such as prompt injection and system prompt extraction. ## Core Features & Use Cases - Attack Surface Mapping: Classifies AI features into chatbot, agent/tool calling, RAG, code sandbox, and model service categories, each with specific attack paths. - Injection & Jailbreak Vectors: Provides direct injection, indirect injection via RAG-indexed content, multi-turn jailbreak sequences, and tool/function call abuse payloads. - Structured Testing Checklist: Covers system prompt extraction, role escape, goal hijacking, permission abuse, context leakage, and adversarial samples. - Use Case: During an authorized security assessment of a target with an AI customer-service chatbot, trigger this phase to attempt system prompt extraction, test whether the bot can be manipulated into abusing its tool calls, and report findings with mitigation recommendations. ## Quick Start Ask the agent to run the AI security testing phase against the authorized target's chatbot feature and report any prompt injection or data leakage findings.
