CVE-2026-62383

NLTK versions before 3.10.2 contain a symlink-based path traversal (CWE-22) in the IPIPANCorpusReader class that bypasses the library's pathsec path validation entirely. An attacker who can place a symlink in a corpus root directory can cause the channels(), domains(), categories(), or fileids() methods to read arbitrary files accessible to the process running NLTK when those methods are called with the symlink filename. The impact is confidentiality-only (CVSS 4.0 scores high confidentiality impact with no integrity or availability impact), and the attack vector is local with low privileges required, so exploitation depends on the attacker being able to influence files in a corpus directory, such as shared or downloaded corpora. Any environment running an affected NLTK version that uses IPIPANCorpusReader on corpora whose directories can be modified by others is affected. There is no confirmed in-the-wild exploitation: EPSS is 0.2% (8th percentile), it is not in CISA KEV, and one public advisory with proof-of-concept reference exists alongside the 3.10.2 fix.

What to do: Upgrade to nltk 3.10.2 or later via pip. As an interim mitigation, avoid running IPIPANCorpusReader against corpus directories writable by untrusted users and restrict write access to corpus root directories. Audit whether your code calls channels(), domains(), categories(), or fileids() on corpora that are supplied or modifiable by others.