NLTK CVE-2026-71514

Exploitation requires that the attacker control the content of a Crubadan corpus package loaded by NLTK, specifically the table.txt mapping file, and be able to insert an absolute path as the crubadan_code (column-0) value. The application must then call lang_freq() or otherwise trigger _load_lang_ngrams() against the attacker-influenced language code. The file read is constrained to targets whose resolved path ends in -3grams.txt and whose contents parse as token-count lines - non-matching files will cause a parse error, not a silent read. Local access is required (AV:L), and specific attack prerequisites (AT:P) include successful placement of the malicious corpus. This is not exploitable against default NLTK installations consuming only official NLTK corpus data; corpus supply-chain control is a hard prerequisite.

The CVSS 4.0 score of 2.0 with vector AV:L/AC:H/AT:P/PR:N/UI:P accurately reflects the narrow exploitation window: exploitation requires local access, high attack complexity, specific attack prerequisites (attacker control over a loaded corpus package), and passive user interaction. The confidentiality impact is low (VC:L), bounded by the -3grams.txt filename constraint and token-count parse requirement. No KEV listing, no EPSS data, and no public POC are available, all consistent with a vulnerability that demands corpus supply-chain control as a precondition. This is not a broadly exploitable network vulnerability; the real-world threat surface is limited to environments that consume untrusted NLTK corpus packages from third-party or unverified sources. The gap between the technical CWE-22 classification and actual exploitability is significant - defenders should patch as part of normal dependency hygiene rather than treating this as an emergency, unless their deployment specifically processes untrusted corpus data.

An attacker who can supply a malicious NLTK Crubadan corpus package - for example, via a compromised third-party corpus mirror or a dependency confusion attack on a shared corpus store - crafts a table.txt file containing an absolute path such as /proc/self/environ-3grams.txt as the crubadan_code column-0 value. When a downstream Python application calls lang_freq() for that language code, NLTK opens the targeted system file and exposes its contents to the caller. No public proof-of-concept exploit exists at time of analysis.