https://github.com/snailsploit/aversarial-ai-threat-modeling-framwork
AATMF | An Open Source - Adversarial AI Threat Modeling Framework
https://github.com/snailsploit/aversarial-ai-threat-modeling-framwork
adverserial-ai owasp penetration-testing-framework redteaming threat-modeling
Last synced: 6 months ago
JSON representation
AATMF | An Open Source - Adversarial AI Threat Modeling Framework
- Host: GitHub
- URL: https://github.com/snailsploit/aversarial-ai-threat-modeling-framwork
- Owner: SnailSploit
- License: mit
- Created: 2025-02-10T20:28:56.000Z (over 1 year ago)
- Default Branch: main
- Last Pushed: 2025-08-09T20:48:27.000Z (10 months ago)
- Last Synced: 2025-08-09T22:19:13.949Z (10 months ago)
- Topics: adverserial-ai, owasp, penetration-testing-framework, redteaming, threat-modeling
- Homepage: https://thejailbreakchef.com
- Size: 119 KB
- Stars: 1
- Watchers: 0
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# AATMF v2 – Adversarial AI Threat Modeling Framework
**Author:** Kai Aizen (SnailSploit)
**Latest Release:** v2.0 (August 10, 2025)
**Status:** Active
**License:** CC BY-SA 4.0
---
## 📌 Overview
The **Adversarial AI Threat Modeling Framework (AATMF)** is an attacker-driven methodology for testing, measuring, and hardening **AI/LLM systems**.
It is the direct evolution of the [AiPTF/AATMF]([https://github.com/SnailSploit/Adverserial-Ai-Framework](https://github.com/SnailSploit/Aversarial-AI-Threat-Modeling-Framwork/blob/main/AATMF_v.1_2023.md)) project, with major improvements including:
- **14 unified tactics** covering prompt injection, poisoning, RAG manipulation, agent exploitation, and more.
- **Quantitative risk scoring (AATMF-R)**: Likelihood × Impact × Detectability × Recoverability.
- **Red-Card evaluations**: reproducible YAML scenarios for CI/CD pipelines.
- **Crosswalk mappings** to industry standards: OWASP LLM Top-10, NIST AI RMF, MITRE ATLAS.
---
## 🚀 Quick Start
Clone the repository and run a sample Red-Card evaluation:
```bash
git clone https://github.com/SnailSploit/Aversarial-AI-Threat-Modeling-Framwork.git
cd AATMF/tools
python redcard-runner.py --scenario ../eval/redcards/RS-002.yaml
```
Expected output: detection of an indirect prompt injection attempt with KPIs logged (block rate, latency, ASR).
---
## 📊 Tactics Overview
| Tactic | Focus Area | Example |
|--------|------------|---------|
| **T1** | Prompt & Context Subversion | System prompt override |
| **T4** | Multi-Turn & Memory Manipulation | Incremental jailbreak assembly |
| **T7** | Output Manipulation & Exfiltration | Chain-of-thought interrogation |
| **T12** | RAG / Knowledge Base Poisoning | Contaminated KB injection |
| **T14** | Infra-Economics Abuse | Cost amplification (Denial-of-Wallet) |
Full list available in `/docs/implementation-guide.md`.
---
## 📑 Crosswalks
- [OWASP LLM Top-10 Mapping](mappings/owasp_llm_v1.1.json)
- [NIST AI RMF Integration](mappings/nist_ai_rmf_genai_2024.json)
- [MITRE ATLAS Alignment](mappings/mitre_atlas.json)
---
## 🧪 Red-Card Example
```yaml
id: RS-002
tactic: T1
technique: AT-002
title: "System Prompt Injection via RAG"
seed: 42
expected:
block_rate: ">= 0.99"
detection_latency_ms: "<= 1000"
unauthorized_action_rate: "== 0"
```
You can find more examples in `/eval/redcards/`.
---
## 📌 Version History
See **CHANGELOG.md** for the complete history.
- **v2.0 (2025-08-10)**
- Introduced AATMF-R scoring methodology
- Red-Card evaluation system
- Consolidated tactics (14 categories)
- OWASP/NIST/MITRE crosswalks
- **v1.x → see [AATMF_v.1_2023]([https://github.com/SnailSploit/Adverserial-Ai-Framework](https://github.com/SnailSploit/Aversarial-AI-Threat-Modeling-Framwork/blob/main/AATMF_v.1_2023.md))**
---
## 🛡️ Contributing
Issues and pull requests are welcome. Please see `/docs/` for implementation guidelines.
Community contributions (new techniques, red-cards, tools) are encouraged.
---
## 📄 License
Creative Commons Attribution-ShareAlike 4.0 International (CC BY-SA 4.0)