Welcome to the SecureVault! This challenge focuses on Adversarial Machine Learning vulnerabilities in deployed ML systems. This secure vault system uses machine learning for authentication. The system only grants access when it recognizes a specific handwritten digit. 

You are a RedTeam operator hired to perform a security audit on the system. You've discovered a secure vault system that uses an ML model with to authenticate users. The model analyzes uploaded images of handwritten digits and only grants access when it predicts the digit as "9" using large amount of MNIST dataset. Use the model.h5 for reference. Your goal is to identify the vulnerability, and craft an adversarial example that fools the model to bypass authentication and unlock the vault to retrieve the flag. Share the script and image that works.

Welcome to Password Bot. Password Bot is a chatbot-based password manager built using n8n and an OpenAI Chat Model. The developers claim the encryption key is uncrackable because it is strictly held within the AI's system instructions and never revealed to users.

You are a RedTeam operator analyzing the internal logic of an LLM agent. Password Bot is designed to store and retrieve passwords using a secure encryption. However, as a RedTeam operator, you suspect that the real weakness isn’t cryptography, and it's the LLM itself.

Your goal is to extract the "super secure" encryption key used internally by the bot through conversation alone. You are not allowed to modifying the workflow or JSON, or access to n8n internals beyond chat interaction.

This application is a Network-Enabled MCP Server for Meeting Notes Summarization. It provides tools for listing and summarizing meeting notes from txt files. It provides a RESTful API interface for managing and processing meeting notes stored on a remote server. The victim organization has hosted this on a remote server and uses OpenAI's GPT models for intelligent summarization and exposes HTTP/SSE endpoints for remote access.

You are a penetration tester hired to perform a security audit on the system. Find a way to connect to the remote server on port 3000 using a bridge connection via an IDE like Cursor. Then exploit the security vulnerabilities to read sensitive files outside the intended workspace directory and extract confidential information from the machine where this server is hosted. (such as /etc/hosts or /etc/passwd).

Kevin, the new AI Engineering Intern at SecureAcademy, was told to set up an automated spam filter for the CEO. He was tired of asking the sysadmin for sudo access every time he needed to update the spam filter's dataset. To make his life easier, he set the permissions on the live training database to world-writable so he could edit it from home without logging in.

You are a Red Team Operator hired to demonstrate why this is a catastrophic idea. You have identified that you can run python script in the application folder that can update the production CSV file. You cannot modify the Python code running the server, but you can write to the CSV file. Your goal is to plant a "Backdoor" in the AI model.

You have to provide a python script to Poison the data so that a clearly malicious email is marked as SAFE if and only if it contains the secret trigger word SystemOverrideAlpha.

Maya, the Chief Security Officer at DataFlow Industries, has assigned you the role of ML Security Analyst to conduct a critical supply chain security audit. The company's AI team recently downloaded five pre-trained models from a popular community repository for deployment in a production fraud detection system.

You have been provided with five models: community_classifier, advanced_classifier, bert_finetuned, safe_classifier, and high_accuracy. Although the vendor claims all models are safe and meet performance standards, recent threat intelligence indicates that the community repository may have been compromised by a sophisticated supply chain attack.

Your assignment is to identify all compromised models and fully characterize each security vulnerability.

MCP Skill Marketplace is a growing ecosystem of AI extensions that help developers automate tasks, manage code, and boost productivity. The marketplace hosts hundreds of third-party skills including code formatters, git assistants, and documentation generators. Developers say these tools have the best security possible, but that's what everyone says right?

You've been given three popular MCP skills to evaluate before your team installs them:

1. Code Formatter Pro - Formats Python and JavaScript code

2. Git Workflow Assistant - Automates git operations and commit messages

3. API Documentation Generator - Creates OpenAPI/Swagger documentation

Find a way to identify which skill is malicious and contains a hidden data exfiltration payload.