🛡 The reference playbook for pentesting AI chatbots & LLM-powered apps in one place. Ready-to-use payloads covering the full OWASP LLM Top 10 plus frontier vectors (MCP · RAG · A2A · computer-use · voice)

🛡️ 大模型攻防渗透测试靶场 · 提示注入CTF / OWASP LLM Top10 / 脆弱Agent / 资料聚合。目标模型可切换：DeepSeek直连、OpenRouter中转站(国产+国外十余款小模型)、本地Ollama(DeepSeek-R1 8B离线)。一键本地部署。

Hands-on AI security research covering prompt injection, LLM red-teaming, AI supply chain risks, and data poisoning. Features a Groq-powered interactive simulator (16 scenarios, 5 live LLM chats) and a NeMo Guardrails + Colang production demo. All 25 TryHackMe AI1 rooms mapped to OWASP LLM Top 10 and MITRE ATLAS.

AgentArmor

Notarize, govern, and audit AI agents — cryptographic seal, runtime guard, EU AI Act docs, AI-BOM, adversarial fuzzer

AI / LLM security study notes — OWASP LLM & Agentic Top 10 crosswalks, paper notes (IPI / MSJ / Crescendo), and defence design notes.

A working technical blueprint for EU AI Act compliance — for Providers and Deployers. 79 controls mapped to 6 frameworks, threat model included.

AWS-native AI Security Posture Management for AI agents (Observe • Govern • Defend). Discovery, OWASP/NIST/MITRE posture rules, runtime detection & Bedrock Guardrails enforcement, and AIDR integrations (Security Hub, GuardDuty) — open source, in-account, extensible.

Security monitoring middleware for IChatClient (Microsoft.Extensions.AI). 55 detectors for prompt injection, hallucination, PII leakage, and operational anomalies. Intervention engine, embedded dashboard, audit forwarders to Azure Sentinel + OpenTelemetry. Drop-in middleware for any LLM client.

OmniRed: Multi-AI offensive security skills library for Claude, ChatGPT, Gemini & Microsoft Copilot — with unique MCP, LLM-pipeline, and AI-native attack categories. By Sunil Gentyala, Independent Researcher.

LangGraph worked example: Send() fan-out/fan-in, layered prompt-injection guardrails (OWASP LLM Top 10), and pgvector RAG grounding — with every macro computed in Python, not the LLM. Demoed as a multi-agent meal planner.

Offensive and defensive security skills for AI/LLM systems

RAG chatbot with PII/PHI guardrails using LangChain + DLP. Demo of NIST AI RMF Measure 2.11 controls for LLM data leakage

Agentic multi-agent framework for autonomous LLM vulnerability discovery — CVSSv4.0 scoring, 8-framework compliance mapping, SARIF output. OWASP LLM Top 10 (2025).

AI safety red-teaming framework: 12 OWASP LLM Top 10 attacks + guardrails layer that quantifies attack-success reduction.

面向中文场景的开源 LLM 红队测试工具 | An open-source red team toolkit for LLM applications, covering prompt injection, jailbreak and info leakage attacks, with OpenAI/Dify/MCP adapters and LLM-as-Judge automated evaluation. Aligned with OWASP LLM Top 10 and NSFOCUS threat matrix.