CVE-2026-62388

NLTK versions before 3.10.0 ship with ENFORCE set to False by default in pathsec.py, so the library's security validation functions only emit warnings instead of raising exceptions. Any application or pipeline that processes untrusted file paths or untrusted pickle data through NLTK's path/deserialization protections can have those protections bypassed by an attacker, since the validations are not actually enforced unless manually enabled. Per the CVSS 4.0 scoring, the primary scored impact is high loss of confidentiality (e.g., accessing files or data outside intended boundaries), with no scored integrity or availability impact. All users of NLTK prior to version 3.10.0 are affected, including academic, research, and production NLP environments. No exploitation has been observed in the wild (not in CISA KEV, EPSS ~0.5%), but one public advisory reference (GHSA-p3m8-78j2-g5p3) is available.

What to do: Upgrade to NLTK 3.10.0 or later, where the security validations are enforced by default. If upgrading is not immediately possible, explicitly enable the ENFORCE setting in pathsec.py for any code that processes untrusted file paths or deserializes untrusted pickles. Review NLP workflows that load external corpora, models, or serialized objects, and ensure they do not rely on the previously warning-only validations.