The vulnerability exists in the ArxivReader class within the llama-index-readers-papers package. Specifically, the load_data and load_papers_and_abstracts methods were found to be vulnerable. The root cause of the vulnerability is the insecure generation of filenames for downloaded Arxiv papers. The filename was created by hashing the title of the paper using the _hacky_hash method. However, paper titles are not guaranteed to be unique. This creates a scenario where two different papers with the same title would result in the same filename. When these papers are downloaded, the second paper would overwrite the first, leading to data loss and preventing the overwritten paper from being processed. The fix, implemented in commit f69e1c0e7579228fec4cfaf716e4f951e131de77, addresses this issue by making the input to the hash function unique. It now combines the paper's title with its unique entry_id before hashing, thus ensuring that each paper gets a unique filename and preventing collisions.