AIKIDO-2026-10445

TL;DR

Data and corpus path helpers in PyThaiNLP joined user-supplied path segments with the intended base directory using plain os.path.join without verifying the resolved path stayed inside that directory. Values containing .. segments could normalize to locations outside the PyThaiNLP data or corpus roots, so downstream file operations could touch unintended paths when callers passed untrusted or externally influenced strings. The fix centralizes containment in safe_path_join, which normalizes paths and rejects resolutions outside the base directory, and removes redundant helper paths that widened the attack surface.