LLM Red Team Web Framework

Large Language Models (LLMs) are increasingly integrated into enterprise systems such as conversational agents, Retrieval-Augmented Generation (RAG) pipelines, autonomous AI agents, and document processing systems. Despite their capabilities, these systems remain vulnerable to prompt injection attacks, jailbreak attempts, sensitive data leakage, insecure tool usage, and unsafe autonomous behavior. Existing testing method-ologies are often fragmented, manual, and lack centralized observability. This paper presents the LLM Red Team Web Framework, a modular AI security testing platform designed to automate adversarial evaluation of LLM-powered systems. The framework supports multiple target types including Chat UI systems, RAG applications, agentic systems, API-based chat systems, and file upload pipelines. The platform enables users to create projects, configure target endpoints, select multiple vulnerability scanners, and execute security tests in isolated Docker-based environments. The framework provides real-time observability dashboards with live attack logs, severity visualization, Attack Success Rate (ASR), leakage metrics, and detection timing analytics. A dedi-cated findings module presents vulnerability evidence, mitigation strategies, and automated JIRA ticket creation support. Experi-mental evaluation demonstrates that the framework successfully automates adversarial testing workflows while improving visibil- ity and reproducibility of AI security assessments.