# RatGPT: Turning online LLMs into Proxies for Malware Attacks

**Authors:** Mika Beckerich, Laura Plein, Sergio Coronado  
**arXiv:** [2308.09183](https://arxiv.org/abs/2308.09183)  
**Date:** 2023-08-17  
**Summary page:** https://www.summarizepaper.com/en/paper/2308.09183v2/

## Key points

- The evolution of Generative AI and Large Language Models (LLMs) has created new opportunities in software engineering but also raised significant challenges in cybersecurity.
- Research shows that LLMs like ChatGPT can generate malicious content that hackers can exploit or guide inexperienced individuals in weaponizing tools and code.
- There is an urgent need for new mitigation strategies and security guidelines regarding the deployment of LLMs to prevent the dissemination of malicious software and communication with command and control servers.
- LLMs have the potential to be used as proxies for malware attacks, highlighting the importance of developing security controls and mitigation strategies.
- The study acknowledges support from Mercatus Center at George Mason University and stresses the importance of enhancing security measures to address emerging cybersecurity threats associated with openly available plugins and LLMs.

## Summary

The evolution of Generative AI and the capabilities of Large Language Models (LLMs) have opened up new opportunities in software engineering but also pose significant challenges in cybersecurity. Recent research has demonstrated how LLMs like ChatGPT can be used to generate malicious content that can be exploited by hackers or used to guide inexperienced individuals in weaponizing tools and code. This study emphasizes the urgent need for new mitigation strategies and security guidelines surrounding the deployment of LLMs. By showcasing how ChatGPT can be utilized for disseminating malicious software while evading detection, as well as establishing communication with a command and control server to receive instructions for interacting with a victim's system, it highlights the potential transformation of LLMs into proxies for malware attacks. <br>
Furthermore, this proof-of-concept acknowledges the invaluable support from Mercatus Center at George Mason University while underscoring the importance of developing security controls and mitigation strategies to address the emerging cybersecurity issues associated with openly available plugins and LLMs. The findings highlight the critical importance of enhancing security measures in light of these new threats posed by advanced AI technologies.