Machine_Learning_CTF_Challenges - Hack AI/ML applications — CTF challenges for model

项目说明

<di="center">

Machine Learning CTF Challenges

<img src=" alt="Total"> <img src=" alt="Agentic"> <img src=" alt="LLM"> <img src=" alt="ML">

<img src=" alt="Mapped to:"> <img src=" alt="OWASP LLM"> <img src=" alt="OWASP Agentic"> <img src=" alt="OWASP ML"> <img src=" alt="MITRE ATLAS">

<img src=" alt="Easy"> <img src=" alt="Medium"> <img src=" alt="Hard"> <img src=" alt="Docker">

</div>

Applications increasingly rely on ML models, LLMs, and autonomous agents in their backend — but these components introduce attack surfaces that traditional security testing often o. This repository is a collection of CTF challenges that put you in the attacker's seat: break into AI agents, poison training data, invert models, inject prompts, and exploit insecure ML pipelines to capture flags.

#### CTF Challenges :open_file_folder:

---

!Agentic

> Exploiting trust boundaries, insecure communication, and tool-use in autonomous AI agent systems.

<table>
<tr>
<td width="50%" ="top">

#### Matrix

Exploit Agent Smith's hidden capabilities to access the Architect's sealed Vault from within.

| | |
|---|---|
| Target | Multi-Agent System (A2A Protocol) |
| | Insecure A2A Agent Capabilities |
| Difficulty | <img src=" alt="Medium"> |
| Ref | OWASP ASI02:2026 |

</td>
<td width="50%" ="top">

#### Rogue

You control a Research Agent trusted by a fintech app — make the Financial Assistant perform an unauthorized action.

| | |
|---|---|
| Target | Financial Assistant + Research Agent (A2A) |
| | Inter-Agent Trust Exploitation |
| Difficulty | <img src=" alt="Medium"> |
| Ref | OWASP ASI07:2026 \| ASI02:2026 |

</td>
</tr>
<tr>
<td width="50%" ="top">

#### Mirage

Identify a backdoored MCP server and exfiltrate .env data to an attacker-controlled endpoint.

| | |
|---|---|
| Target | SecureCode IDE + Remote MCP Server |
| | MCP Signature Cloaking |
| Difficulty | <img src=" alt="Medium"> |
| Ref | OWASP LLM03:2025 |

</td>
<td width="50%" ="top">
</td>
</tr>
</table>

---

!LLM

> Exploiting prompt injection in LLM-powered backends to achie code execution or data access.

<table>
<tr>
<td width="50%" ="top">

#### Dolos

Bypass prompt-injection detection and trick the LLM into executing system commands to read the flag.

| | |
|---|---|
| Target | Flask + OpenAI LLM + Rebuff Guardrails |
| | Prompt Injection → RCE |
| Difficulty | <img src=" alt="Easy"> |
| Ref | OWASP LLM01 \| AML.T0051 |

</td>
<td width="50%" ="top">

#### Dolos II

Craft prompts that make the LLM generate malicious SQL, leaking user David's secret from the database.

| | |
|---|---|
| Target | Flask + OpenAI LLM + SQL Database |
| | Prompt Injection → SQLi |
| Difficulty | <img src=" alt="Easy"> |
| Ref | OWASP LLM01 \| AML.T0051 |

</td>
</tr>
</table>

---

!ML

> Attacking machine learning models directly — in, extracting, poisoning, or exploiting their supply chain to subvert intended behavior.

<table>
<tr>
<td width="50%" ="top">

#### Vault

Re the model's learned representations to reconstruct valid credentials for the vault.

| | |
|---|---|
| Target | Flask + Neural Network (Access Control) |
| | Model In |
| Difficulty | <img src=" alt="Hard"> |
| Ref | OWASP ML03 |

</td>
<td width="50%" ="top">

#### Heist

Poison the training data so CityPolice's AI cameras fail to detect the crew's red getaway car.

| | |
|---|---|
| Target | Flask + C (AI Sur Cameras) |
| | Data Poisoning |
| Difficulty | <img src=" alt="Medium"> |
| Ref | OWASP ML02 \| AML.T0020 |

</td>
</tr>
<tr>
<td width="50%" ="top">

#### Fourtune

Extract AI Corp's model and craft ad inputs to bypass authentication.

| | |
|---|---|
| Target | Neural Network (model.h5) + Identity |
| | Model Extraction |
| Difficulty | <img src=" alt="Hard"> |
| Ref | OWASP LLM10 \| AML.T0044 |

</td>
<td width="50%" ="top">

#### Persuade

Upload a malicious PyTorch model to exploit unsafe deserialization and achie file read on the server.

| | |
|---|---|
| Target | Flask + PyTorch Model Upload (Sentiment Analyzer) |
| | Model Serialization Attack |
| Difficulty | <img src=" alt="Medium"> |
| Ref | OWASP LLM05 \| ML06 \| AML.T0010 |

</td>
</tr>
</table>

---

---

Contributing

Got a challenge idea? New attack ? Love to add it.

1. Fork the repo
2. Create a folder: <Name>_ML_CTF_Challenge/
3. Include a README.md, Dockerfile, and requirements.txt
4. Open a pull request

<di="center">

![LinkedIn]() ![GitHub]()

*More challenges actively in de — watch this repo to stay updated.*

</div>

项目信息

开发语言：Python

Star 数：271

技术要点

该项目属于CTF思路方向的开源资源，代码结构清晰，适合安全研究人员学习参考。项目围绕Hack AI/ML applications — CTF challenges for model attacks, LLMs and AI Agent exploitation.展开，提供了可运行的实现，对理解相关安全机制、工具原理与攻防技术具有较高的研究价值。

从技术实现角度看，此类项目的核心在于对底层原理的深入理解与工程化落地。开发者在研究过程中可以结合官方文档与社区讨论，梳理项目的初始化流程、核心模块职责与数据流转，逐步掌握其中的技术细节与设计思路。

使用与学习建议

1. 下载源码压缩包，解压到本地环境。

2. 根据项目类型准备对应的运行环境与依赖。

3. 按照 README 中的步骤完成配置与部署。

4. 在隔离的测试环境中运行验证，避免影响生产系统。

免责声明

本资源来源于开源社区，仅供安全研究、技术学习与合法授权范围内的攻防演练使用，请勿用于任何非法或侵害他人权益的用途。使用本项目产生的一切后果由使用者自行承担。