CVE-2024-10954
📋 TL;DR
This vulnerability allows remote code execution on servers running vulnerable versions of the gpt_academic manim plugin. Attackers can inject malicious code through LLM prompts, which gets executed without proper sandboxing. Anyone using gpt_academic with the manim plugin before the fix is affected.
💻 Affected Systems
- binary-husky/gpt_academic with manim plugin
📦 What is this software?
Gpt Academic by Binary Husky
⚠️ Risk & Real-World Impact
Worst Case
Full compromise of the backend server, allowing attackers to execute arbitrary commands, steal data, install malware, or pivot to other systems.
Likely Case
Attacker gains shell access to the application server, potentially accessing sensitive data and disrupting services.
If Mitigated
With proper input validation and sandboxing, the risk is reduced to minimal, though prompt injection attempts may still occur.
🎯 Exploit Status
Requires ability to submit prompts to the LLM interface and knowledge of prompt injection techniques.
🛠️ Fix & Mitigation
✅ Official Fix
Patch Version: Latest version after the fix
Vendor Advisory: https://huntr.com/bounties/72d034e3-6ca2-495d-98a7-ac9565588c09
Restart Required: No
Instructions:
1. Update gpt_academic to the latest version. 2. Ensure the manim plugin is updated. 3. Verify sandboxing is properly implemented.
🔧 Temporary Workarounds
Disable manim plugin
allTemporarily disable the vulnerable manim plugin to prevent exploitation.
Remove or comment out manim plugin loading in configuration
Implement input validation
allAdd strict input validation to reject suspicious prompts containing code execution patterns.
🧯 If You Can't Patch
- Implement network segmentation to isolate the vulnerable system
- Deploy application-level firewall with WAF rules to detect prompt injection attempts
🔍 How to Verify
Check if Vulnerable:
Check if running a version before the fix and if manim plugin executes LLM-generated code without sandboxing.Check Version:
Check gpt_academic version in configuration or package managerVerify Fix Applied:
Verify the latest version is installed and test that malicious prompts no longer execute arbitrary code.📡 Detection & Monitoring
Log Indicators:
- Unusual command execution patterns
- Suspicious prompt submissions
- Error logs from sandbox violations
Network Indicators:
- Unexpected outbound connections from the application server
- Unusual traffic patterns to/from LLM interface
SIEM Query:
Search for 'manim plugin execution' or 'LLM code execution' events with suspicious patterns