VectifyAI PageIndex PDF Table of Contents Handler Denial-of-Service Vulnerability

A denial-of-service vulnerability has been identified in VectifyAI PageIndex versions prior to f50e52975313c6716c02b20a119577a1929decba. The issue arises in the PDF Table of Contents Handler component, specifically within the 'toc_transformer' function of 'pageindex/page_index.py'. This vulnerability allows for an infinite loop to be created, which can be exploited remotely. The loop is triggered by a crafted PDF that contains an excessively long Table of Contents, causing the application to repeatedly query the backend LLM API without completion. This not only exhausts processing resources but also rapidly depletes LLM API credits, leading to financial costs.

Impact

Exploitation of this vulnerability causes an infinite loop that consumes 100% of a CPU thread, while continuously sending requests to the LLM API. This drains both computational resources and financial credits associated with the API usage, causing a denial-of-service condition on the affected system.

Reproduction

To reproduce this vulnerability, generate a PDF file that includes thousands of sections in the Table of Contents, ensuring it is long enough to cause the LLM to truncate its output. Alternatively, set up a Mock OpenAI proxy that returns a 'length' finish reason, simulating the truncation. Then, run the PageIndex application via the command line, specifying the path to the malicious PDF and the model to be used. The application will become stuck in an infinite loop, processing the PDF and depleting API credits.