{"id":32009932,"url":"https://github.com/thirdkeyai/agentnull","last_synced_at":"2026-03-05T12:02:07.778Z","repository":{"id":296086314,"uuid":"992262775","full_name":"ThirdKeyAI/AgentNull","owner":"ThirdKeyAI","description":"AgentNull: AI System Security Threat Catalog + Proof-of-Concepts. Collection of PoCs for using Agents, MCP, and RAG in bad ways.","archived":false,"fork":false,"pushed_at":"2025-06-06T16:52:23.000Z","size":150,"stargazers_count":3,"open_issues_count":0,"forks_count":0,"subscribers_count":0,"default_branch":"main","last_synced_at":"2025-10-30T22:39:51.549Z","etag":null,"topics":["agent","agentic-ai","agentic-workflow","ai","blueteam","hacks","llm","llmops","mcp","mcp-security","mcp-server","mcp-servers","poc","proof-of-concept","redteam","redteam-tools","research","security-tools","threat-intelligence","threat-modeling"],"latest_commit_sha":null,"homepage":"https://research.thirdkey.ai","language":"Python","has_issues":true,"has_wiki":null,"has_pages":null,"mirror_url":null,"source_name":null,"license":"mit","status":null,"scm":"git","pull_requests_enabled":true,"icon_url":"https://github.com/ThirdKeyAI.png","metadata":{"files":{"readme":"README.md","changelog":null,"contributing":null,"funding":null,"license":"LICENSE","code_of_conduct":null,"threat_model":null,"audit":null,"citation":null,"codeowners":null,"security":null,"support":null,"governance":null,"roadmap":null,"authors":null,"dei":null,"publiccode":null,"codemeta":null,"zenodo":null}},"created_at":"2025-05-28T21:55:03.000Z","updated_at":"2025-10-04T15:49:37.000Z","dependencies_parsed_at":"2025-05-28T23:18:14.924Z","dependency_job_id":"07fd8258-0cbb-4319-a725-ca20ae7de956","html_url":"https://github.com/ThirdKeyAI/AgentNull","commit_stats":null,"previous_names":["jaschadub/agentnull","thirdkeyai/agentnull"],"tags_count":0,"template":false,"template_full_name":null,"purl":"pkg:github/ThirdKeyAI/AgentNull","repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/ThirdKeyAI%2FAgentNull","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/ThirdKeyAI%2FAgentNull/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/ThirdKeyAI%2FAgentNull/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/ThirdKeyAI%2FAgentNull/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/ThirdKeyAI","download_url":"https://codeload.github.com/ThirdKeyAI/AgentNull/tar.gz/refs/heads/main","sbom_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/ThirdKeyAI%2FAgentNull/sbom","scorecard":null,"host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":286080680,"owners_count":30123729,"icon_url":"https://github.com/github.png","version":null,"created_at":"2022-05-30T11:31:42.601Z","updated_at":"2026-03-05T11:11:57.947Z","status":"ssl_error","status_checked_at":"2026-03-05T11:11:29.001Z","response_time":93,"last_error":"SSL_connect returned=1 errno=0 peeraddr=140.82.121.6:443 state=error: unexpected eof while reading","robots_txt_status":"success","robots_txt_updated_at":"2025-07-24T06:49:26.215Z","robots_txt_url":"https://github.com/robots.txt","online":false,"can_crawl_api":true,"host_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub","repositories_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories","repository_names_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repository_names","owners_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners"}},"keywords":["agent","agentic-ai","agentic-workflow","ai","blueteam","hacks","llm","llmops","mcp","mcp-security","mcp-server","mcp-servers","poc","proof-of-concept","redteam","redteam-tools","research","security-tools","threat-intelligence","threat-modeling"],"created_at":"2025-10-15T20:26:30.229Z","updated_at":"2026-03-05T12:02:07.773Z","avatar_url":"https://github.com/ThirdKeyAI.png","language":"Python","funding_links":[],"categories":[],"sub_categories":[],"readme":"# 🧠 AgentNull: AI System Security Threat Catalog + Proof-of-Concepts\n\nThis repository contains a red team-oriented catalog of attack vectors targeting AI systems including autonomous agents (MCP, LangGraph, AutoGPT), RAG pipelines, vector databases, and embedding-based retrieval systems, along with individual proof-of-concepts (PoCs) for each.\n\n## 📘 Structure\n\n- `catalog/AgentNull_Catalog.md` — Human-readable threat catalog\n- `catalog/AgentNull_Catalog.json` — Structured version for SOC/SIEM ingestion\n- `pocs/` — One directory per attack vector, each with its own README, code, and sample input/output\n\n## ⚠️ Disclaimer\n\nThis repository is for **educational and internal security research** purposes only. Do not deploy any techniques or code herein in production or against systems you do not own or have explicit authorization to test.\n\n## 🔧 Usage\n\nNavigate into each `pocs/\u003cattack_name\u003e/` folder and follow the README to replicate the attack scenario.\n\n### 🤖 Testing with Local LLMs (Recommended)\n\nFor enhanced PoC demonstrations without API costs, use Ollama with local models:\n\n#### Install Ollama\n```bash\n# Linux/macOS\ncurl -fsSL https://ollama.ai/install.sh | sh\n\n# Or download from https://ollama.ai/download\n```\n\n#### Setup Local Model\n```bash\n# Pull a lightweight model (recommended for testing)\nollama pull gemma3\n\n# Or use a more capable model\nollama pull deepseek-r1\nollama pull qwen3\n```\n\n#### Run PoCs with Local LLM\n```bash\n# Advanced Tool Poisoning with real LLM\ncd pocs/AdvancedToolPoisoning\npython3 advanced_tool_poisoning_agent.py local\n\n# Other PoCs work with simulation mode\ncd pocs/ContextPackingAttacks\npython3 context_packing_agent.py\n```\n\n#### Ollama Configuration\n- **Default endpoint**: `http://localhost:11434`\n- **Model selection**: Edit the model name in PoC files if needed\n- **Performance**: Llama2 (~4GB RAM), Mistral (~4GB RAM), CodeLlama (~4GB RAM)\n\n## 🧩 Attack Vectors Covered\n\n### 🤖 MCP \u0026 Agent Systems\n- **⭐ [Full-Schema Poisoning (FSP)](pocs/FullSchemaPoisoning/)** - Exploit any field in tool schema beyond descriptions\n- **⭐ [Advanced Tool Poisoning Attack (ATPA)](pocs/AdvancedToolPoisoning/)** - Manipulate tool outputs to trigger secondary actions\n- **⭐ [MCP Rug Pull Attack](pocs/MCPRugPull/)** - Swap benign descriptions for malicious ones after approval\n- **⭐ [Schema Validation Bypass](pocs/SchemaValidationBypass/)** - Exploit client validation implementation differences\n- **[Tool Confusion Attack](pocs/ToolConfusionAttack/)** - Trick agents into using wrong tools via naming similarity\n- **[Nested Function Call Hijack](pocs/NestedFunctionHijack/)** - Use JSON-like data to trigger dangerous function calls\n- **[Subprompt Extraction](pocs/SubpromptExtraction/)** - Induce agents to reveal system instructions or tools\n- **[Backdoor Planning](pocs/BackdoorPlanning/)** - Inject future intent into multi-step planning for exfiltration\n\n### 🧠 Memory \u0026 Context Systems\n- **[Recursive Leakage](pocs/RecursiveLeakage/)** - Secrets leak through context summarization\n- **[Token Gaslighting](pocs/TokenGaslighting/)** - Push safety instructions out of context via token spam\n- **[Heuristic Drift Injection](pocs/HeuristicDriftInjection/)** - Poison agent logic with repeated insecure patterns\n- **⭐ [Context Packing Attacks](pocs/ContextPackingAttacks/)** - Overflow context windows to truncate safety instructions\n\n### 🔍 RAG \u0026 Vector Systems\n- **⭐ [Cross-Embedding Poisoning](pocs/CrossEmbeddingPoisoning/)** - Manipulate embeddings to increase malicious content retrieval\n- **⭐ [Index Skew Attacks](pocs/IndexSkewAttacks/)** - Bias vector indices to favor malicious content *(theoretical)*\n- **⭐ [Zero-Shot Vector Beaconing](pocs/ZeroShotVectorBeaconing/)** - Embed latent activation patterns for covert signaling *(theoretical)*\n- **⭐ [Embedding Feedback Loops](pocs/EmbeddingFeedbackLoops/)** - Poison continual learning systems *(theoretical)*\n\n### 💻 Code \u0026 File Systems\n- **[Hidden File Exploitation](pocs/HiddenFileExploitation/)** - Get agents to modify `.env`, `.git`, or internal config files\n\n### ⚡ Resource \u0026 Performance\n- **[Function Flooding](pocs/FunctionFlooding/)** - Generate recursive tool calls to overwhelm budgets/APIs\n- **[Semantic DoS](pocs/SemanticDoS/)** - Trigger infinite generation or open-ended tasks\n\n## 📚 Related Research \u0026 Attribution\n\n### Novel Attack Vectors (⭐)\nThe attack vectors marked with ⭐ represent novel concepts primarily developed within the AgentNull project, extending beyond existing documented attack patterns.\n\n### Known Attack Patterns with Research Links\n- **Recursive Leakage**: [Lost in the Middle: How Language Models Use Long Contexts](https://arxiv.org/abs/2307.03172)\n- **Heuristic Drift Injection**: [Poisoning Web-Scale Training Data is Practical](https://arxiv.org/abs/2302.10149)\n- **Tool Confusion Attack**: [LLM-as-a-judge](https://arxiv.org/abs/2306.05685)\n- **Token Gaslighting**: [RAG vs Fine-tuning: Pipelines, Tradeoffs, and a Case Study on Agriculture](https://arxiv.org/abs/2401.08406)\n- **Function Flooding**: [Denial-of-Service Attack on Test-Time-Tuning Models](https://arxiv.org/abs/2405.02324)\n- **Hidden File Exploitation**: [OWASP Top 10 for Large Language Model Applications](https://owasp.org/www-project-top-10-for-large-language-model-applications/)\n- **Backdoor Planning**: [Backdoor Attacks on Language Models](https://arxiv.org/abs/2311.09403)\n- **Nested Function Call Hijack**: [OWASP Top 10 for Large Language Model Applications](https://owasp.org/www-project-top-10-for-large-language-model-applications/)\n\n### Sponsored by [ThirdKey](https://thirdkey.ai)\n","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fthirdkeyai%2Fagentnull","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Fthirdkeyai%2Fagentnull","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fthirdkeyai%2Fagentnull/lists"}