Hash Collision Vulnerability in Run-Llama's Document Reader

CVE-2025-6211

What is CVE-2025-6211?

A vulnerability exists in the DocugamiReader class of the run-llama/llama_index repository prior to version 0.12.28, where the use of MD5 hashing for generating document chunk IDs can lead to hash collisions. This can result in structurally distinct document chunks that share identical text overwriting each other. Consequently, critical and legally significant content may be lost, parent-child chunk hierarchies could be disrupted, and AI-generated outputs might become inaccurate or fabricative. The issue has been addressed in version 0.3.1.

References