# A First Look at GPT Apps: Landscape and Vulnerability

**Authors:** Zejun Zhang, Li Zhang, Xin Yuan, Anlan Zhang, Mengwei Xu, Feng Qian  
**arXiv:** [2402.15105](https://arxiv.org/abs/2402.15105)  
**Date:** 2024-02-23  
**Summary page:** https://www.summarizepaper.com/en/paper/2402.15105v2/

## Key points

- Dedicated LLM app stores have emerged in response to the proliferation of GPT apps, such as \textit{GPTStore.AI} and the official \textit{OpenAI GPT Store}.
- There is a significant lack of understanding regarding the evolution and vulnerabilities of this new ecosystem.
- A comprehensive longitudinal study spanning over 5 months was conducted to analyze the emerging LLM app ecosystem by examining metadata and user feedback for all GPT apps across these two prominent stores.
- Configurations, including system prompts, knowledge files, and APIs for the top 10,000 popular apps were analyzed using automated tools and a TriLevel configuration extraction strategy.
- Findings reveal an increase in user enthusiasm for GPT apps over time but a plateau in creator interest within three months of launch.
- Nearly 90% of system prompts can be easily accessed due to inadequate security measures, leading to instances of plagiarism and duplication among apps within the ecosystem.
- Urgent enhancements are needed within the LLM app ecosystem by stakeholders to ensure sustainable growth and development.

## Summary

In response to the proliferation of GPT apps following OpenAI's introduction of GPTs, dedicated LLM app stores have emerged. These include \textit{GPTStore.AI} and the official \textit{OpenAI GPT Store}. However, despite the rapid growth of this new ecosystem, there remains a significant lack of understanding regarding its evolution and vulnerabilities. To address this gap in knowledge, this paper presents a comprehensive longitudinal study spanning over 5 months. The study focuses on analyzing the emerging LLM app ecosystem by examining metadata (such as names, creators, descriptions) and user feedback for all GPT apps across these two prominent stores. Additionally, configurations including system prompts, knowledge files and APIs for the top 10 000 popular apps are also analyzed using automated tools and a TriLevel configuration extraction strategy. The findings from this extensive analysis reveal intriguing trends within the LLM app ecosystem. Firstly there is a noticeable increase in user enthusiasm for GPT apps over time. Conversely creator interest appears to plateau within three months of the initial launch of GPTs. Furthermore a concerning discovery indicates that nearly 90% of system prompts can be easily accessed due to inadequate security measures in place for GPT app configurations. This vulnerability has led to instances of plagiarism and duplication among apps within the ecosystem. Overall these findings underscore the urgent need for enhancements within the LLM app ecosystem by all stakeholders involved - including app stores creators and users. By addressing these vulnerabilities and fostering a more secure environment for innovation and creativity with GPT apps it is possible to ensure sustainable growth and development within this rapidly evolving landscape.