
Loading

Loading
We use strictly necessary cookies to run this site, and analytics cookies to understand how it's used. See our Privacy Policy for details.
A GitHub-reviewed security advisory, not necessarily under active exploitation the way our CISA KEV advisories are, but a real, disclosed vulnerability in an open-source dependency.
Affected packages
Vulnerable: <= 3.10.3
Patched in 3.10.4
Details
Path traversal in NLTK's save_maxent_params() and load_maxent_params() allows arbitrary file write and arbitrary file read outside the directories permitted by NLTK's pathsec security mechanism.
This vulnerability was previously reported through Huntr, accepted by the NLTK project, and assigned CVE-2026-15367. A $750 bounty was paid through Huntr (i didnt get the fix bounty). Huntr subsequently shut down its OSS vulnerability program, so I am submitting this report through GitHub to continue the existing disclosure process and preserve the original CVE assignment and researcher credit.
The affected functions are:
Both functions construct filesystem paths from the caller-controlled tab_dir argument and access them using Python's built-in open() without passing the paths through NLTK's pathsec validation.
For example, load_maxent_params() constructs and opens:
Similarly, save_maxent_params() creates/writes files under the supplied tab_dir.
This bypasses the security boundary provided by nltk.pathsec. NLTK documents pathsec as the mechanism for restricting filesystem access to permitted roots, with strict enforcement intended to prevent unauthorized filesystem access.
The relevant implementation is in:
nltk/classify/maxent.py
The affected functions can be found in the model parameter loading/saving section of that file.
This issue was previously disclosed through Huntr and accepted before the Huntr OSS program was shut down. The original Huntr report was assigned CVE-2026-15367 and received a $750 bounty.
I am therefore submitting this GitHub Security Advisory primarily as a continuation/transfer of the existing disclosure rather than as a new vulnerability report.
The issue can be reproduced by enabling NLTK's path security enforcement and supplying a tab_dir outside the permitted root.
For example, with a controlled test environment:
import os
import tempfile
import nltk.pathsec as pathsec
from nltk.classify.maxent import load_maxent_params, save_maxent_params
pathsec.ENFORCE = True
allowed_dir = tempfile.mkdtemp()
outside_dir = tempfile.mkdtemp()
# The same outside-root path is rejected when accessed through pathsec.
try:
with pathsec.open(os.path.join(outside_dir, "test.txt"), "w") as f:
f.write("test")
except Exception as e:
print("pathsec.open rejected outside-root access:", type(e).__name__)
# The vulnerable model-parameter APIs directly use open()
# on paths derived from tab_dir.
#
# Supplying outside_dir as tab_dir causes the functions to access
# files under outside_dir rather than enforcing the pathsec boundary.
For save_maxent_params(), the attacker-controlled tab_dir determines where the model parameter files are written.
For load_maxent_params(), the attacker-controlled tab_dir determines which files are read.
The important security property being bypassed is that pathsec.ENFORCE = True does not protect these file operations because the affected functions call the built-in open() directly rather than the pathsec-aware file-opening mechanism.
This is a path traversal / filesystem sandbox bypass resulting in arbitrary local file read and arbitrary local file write.
Applications that use NLTK's pathsec mechanism as a filesystem security boundary may incorrectly assume that model parameter loading and saving are restricted to approved directories.
An attacker who can influence the tab_dir argument may therefore:
The practical impact depends on how an application exposes these APIs and what filesystem permissions the NLTK process has.
CVE: CVE-2026-15367
This vulnerability was previously reported and accepted through Huntr before the Huntr OSS vulnerability program was shut down. The report received a $750 bounty and a CVE was reserved.
I am submitting this report because the original Huntr disclosure workflow is no longer available and the CVE has not yet been published.
Please associate this report with CVE-2026-15367 rather than assigning a new CVE, unless the NLTK maintainers determine that the existing CVE does not correspond to this issue.
Status: MERGED on develop. The load side is in every release since 3.9.4 (including v3.10.3); the save side is on develop and unreleased.
Fix commits: f2556f9e1 (load_maxent_params reads through open_datafile on a PathPointer whose open is pathsec.open; contained in tag 3.9.4 and every later tag) and 2a92b7182 (save_maxent_params validates tab_dir and writes every parameter file through pathsec.open), PR #3759 "Harden maxent / named-entity model-artifact saves against shared-tmp planting", merged 2026-08-18, merge commit a167e94bc on develop.
Probe: nltk/test/unit/security_probes/ghsa_59f9_gqg8_mqpj.py, added on the open PR #3753 (head da55cab0e at the time of this note; measured on its earlier head 238f81500) together with commit 15f84f036 (load_maxent_params accepts a plain str or Path through FileSystemPathPointer). Verdict on 238f81500: FIXED (save and load outside every data root refused by pathsec). The same probe run against develop 2fa3f48fe reports the save refused and a PathPointer load refused, while a str or Path tab_dir raises AttributeError before any file is opened (the probe counts that crash as not fixed; no file is read or written).
EPSS, exploit probability
Low0.27%
estimated chance of real-world exploitation in the next 30 days, higher than 19.8% of every CVE FIRST.org scores
Refreshed 11/16/2026, via FIRST.org's EPSS model, not CVSS, this measures likelihood of exploitation, not how severe it would be.
More pip advisories
Talk to us about this
Find out whether this dependency (or others like it) is actually reachable in your codebase.
Get help patching, or building dependency-scanning into your pipeline going forward.
A secure code review or supply chain review finds this before an advisory does.