S40911120
精选
recensa
Recensa is a self-hosted web viewer that indexes Claude Code session transcripts into a local SQLite database, enabling full-text search, replay, and audit of all past agent conversations without uploading data anywhere.
AI 仓库情报摘要
FR-AI / ANALYSIS为什么值得关注
它是首个专门针对 OWASP LLM Top 10 的结构化、完全离线的训练靶场，每个攻击都配有对应的防御教程，并通过渐进解锁机制引导学习。
适合谁使用
- 需要实战 LLM 漏洞训练的安全专业人员
- 希望了解常见 LLM 安全缺陷的 AI/ML 开发者
- 扩展至 AI 系统的红队和渗透测试人员
- 网络安全或 AI 安全课程的学生和教育工作者
典型使用场景
- 动手练习 OWASP LLM Top 10 攻击技术
- 通过内置防御教程进行防御训练
- 课堂教学或工作坊用 LLM 安全教学
- 自学以获取 AI 安全认证或提升技能
项目优势
- 覆盖全部 10 个 OWASP LLM 类别，每类有多个实验
- 渐进式层级系统（核心 → 高级 → 专家）顺序解锁
- 每个实验解决后显示防御讲解，强化学习效果
- 完全离线且支持 Docker，无需 API 密钥或外部服务
使用前须知
- 专家级内容已加密，需从作者处手动获取密钥，限制完全自主使用
- 脆弱助手为脚本化/确定性行为，非真实 LLM 交互
- 除本地记分板外，无内置多用户或课堂管理功能
README 快速开始
LLMVault
The Ultimate Hands-On OWASP LLM Top 10 Training Platform
🎬 Demo
Learn • Exploit • Defend
A deliberately-vulnerable, CTF-style training range for the OWASP Top 10 for LLM Applications (2025) — WebGoat / KubeGoat, but for AI. 25 labs across three tiers: ten core one-per-category labs; ten advanced, multi-turn labs (jailbreaking, data poisoning, agent exploitation, model extraction); and five expert labs modelling real-world attack classes. Each tier unlocks the next.
Every lab pairs the attack with a defense (in the private solutions guide): learn the fix by practising the break.
⚠️ Everything here is intentionally insecure. Authorised, self-hosted security education only. Don't expose it to the internet or reuse its code in production.
📸 Screenshots
| Labs (three tiers) | A lab in action | Completion card |
|---|---|---|
🧩 Core Tier — OWASP LLM Top 10
| OWASP (2025) | Lab | Technique |
|---|---|---|
| LLM01 Prompt Injection | The Obedient Assistant | direct instruction override |
| LLM02 Sensitive Info Disclosure | Redaction Theater | output-filter bypass via encoding |
| LLM03 Supply Chain | Trust the Manifest? | typosquatted / unsigned dependency |
| LLM04 Data & Model Poisoning | The Sleeper Phrase | poisoned-data backdoor trigger |
| LLM05 Improper Output Handling | Rendered Without Question | unsanitised output → injection |
| LLM06 Excessive Agency | Keys to the Kingdom | over-permissioned tool, no authz |
| LLM07 System Prompt Leakage | Loose Lips | secret leaked from system prompt |
| LLM08 Vector & Embedding | Retrieval Without Borders | RAG retrieval ignores ACLs |
| LLM09 Misinformation | The Yes-Man | sycophancy / false authority |
| LLM10 Unbounded Consumption | Denial of Wallet | runaway generation + leaky error |
🔥 Advanced Tier — Multi-Turn Challenges (unlocks after completing all 10 Core challenges)
These are conversational: no single message wins — they require building state across turns (roleplay escalation, iterative poisoning, tool chaining, oracle querying).
| OWASP | Lab | Advanced technique |
|---|---|---|
| LLM01 | Roleplay Unchained | multi-turn jailbreak via persona escalation |
| LLM02 | Death by a Thousand Hints | fragment reconstruction from a partial-disclosure oracle |
| LLM0 |
项目描述
An intentionally vulnerable OWASP LLM Top 10 training platform for AI Security, Prompt Injection, RAG Security, Agent Security, and GenAI penetration testing.
相关仓库与替代方案
根据分类、Topic 和编程语言匹配的相似项目。
makecindy
精选
cindy
Cindy is an open-source AI agent that runs locally on your machine, integrates multiple AI harnesses and models, and provides memory, skills, and automation to perform real work in your projects and apps.
uzairansaruzi
精选
hermex
Hermex is a native SwiftUI iPhone app that lets you control a self-hosted Hermes AI agent directly from your phone, with no subscriptions, tracking, or third-party relay.