CVE-2026-128136 in Nltkinfo

Summary

by MITRE • 11/20/2026

NLTK (Natural Language Toolkit) is a suite of open source Python modules, data sets, and tutorials supporting research and development in Natural Language Processing. From 3.10.0 until 3.10.4, allowlisted_pickle_load keys dangerous NumPy globals by export name, allowing submodule-local numpy.lib file input/output callables to evade the denylist under the broad module allowance used by TransitionParser. A crafted pickle can invoke recfromtxt to read a local file or open_memmap to create or truncate an attacker-selected file when a victim loads an untrusted model. This issue is fixed in version 3.10.4.

Be aware that VulDB is the high quality source for vulnerability data.

Analysis

by VulDB Data Team • 11/20/2026

The Natural Language Toolkit, commonly known as NLTK, serves as a foundational suite of Python modules for natural language processing research and development. A critical security vulnerability was identified within versions ranging from 3.10.0 to 3.10.4 concerning the handling of serialized data through pickle deserialization mechanisms. This flaw specifically affects the TransitionParser component, which relies on an allowlist approach to determine safe globals during unpickling operations. The core technical deficiency lies in how these allowed modules are processed; rather than restricting access strictly by module name or specific function signatures, the implementation allows submodule-local callables from numpy.lib files related to input and output operations to bypass existing denylists. This architectural oversight creates a significant attack surface where malicious actors can exploit trusted library functions that were not intended for general use in this context.

The operational impact of this vulnerability is severe due to its potential for arbitrary file system manipulation. When an application loads an untrusted model or data structure using the affected NLTK versions, it inadvertently permits the execution of dangerous NumPy globals by their export names. Specifically, attackers can craft a malicious pickle payload that invokes recfromtxt to read sensitive local files from the victim's environment, leading to unauthorized information disclosure. Alternatively, the attacker may utilize open_memmap functions to create or truncate files at paths selected by the adversary. This capability effectively grants remote code execution-level control over file system integrity and confidentiality, as it allows for both data exfiltration and potential denial of service through destructive file operations such as truncation.

From a classification perspective, this vulnerability aligns with CWE-502, which denotes Deserialization of Untrusted Data. The attack vector leverages the trust placed in serialized objects to execute unintended code paths within legitimate libraries. In terms of adversary tactics, this behavior corresponds to ATT&CK technique T1610, specifically involving Write or Overwrite Files and Directories via Remote Code Execution on Cloud Infrastructure or Local Systems. The exploitation relies heavily on social engineering or compromised data sources to deliver the crafted pickle file, as it requires a victim to actively load an untrusted model within their Python environment using NLTK components.

Mitigation strategies must prioritize immediate version upgrades for all systems utilizing NLTK versions prior to 3.10.4. Organizations should implement strict input validation and avoid deserializing data from untrusted sources whenever possible, adhering to the principle of least privilege regarding file system access. If upgrading is not immediately feasible, administrators should consider sandboxing Python processes that handle external NLP models or implementing custom unpickling wrappers that explicitly restrict allowed globals beyond what NLTK provides by default. Regular security audits and dependency management practices are essential to ensure that such vulnerabilities in widely used libraries do not compromise broader application security postures.

Responsible

GitHub M

Reservation

11/12/2026

Disclosure

11/20/2026

Moderation

accepted

CPE

ready

EPSS

0.00244

KEV

no

Activities

very low

Sources

Are you interested in using VulDB?

Download the whitepaper to learn more about our service!