NLTK CVE-2026-78681

Exploitation requires that attacker-controlled or untrusted XML data reaches one of NLTK's internal modules that invoke xml.etree.ElementTree for parsing - versions prior to 3.10.3 only. The CVSS 4.0 vector (PR:N/UI:N/AV:N/AC:L/AT:N) confirms no authentication, no user interaction, and no special attack preconditions are needed once the XML input pathway is reachable. The primary limiting factor is deployment context: NLTK is overwhelmingly used as an offline library in research and data processing pipelines rather than as a network-exposed service, so the XML-parsing paths are frequently never reachable from an untrusted network. Deployments that wrap NLTK in a web API, process third-party XML feeds, or accept user-uploaded corpus files are at materially higher risk. No platform-specific configuration is required to be vulnerable; the flaw is in the default parsing behavior across all supported platforms.

The CVSS 4.0 vector AV:N/AC:L/AT:N/PR:N/UI:N/VA:H yields a score of 8.7, reflecting a remotely exploitable, unauthenticated, low-complexity DoS with high availability impact on the vulnerable system. The attack technique is generic and well-understood, requiring no vulnerability-specific research to weaponize. However, real-world risk is moderated by NLTK's typical deployment profile: it is primarily a library used in NLP research pipelines, data science notebooks, and batch processing workflows - not a public-facing network service. Exploitation therefore requires an application to pipe untrusted or attacker-controlled XML through NLTK's affected parsing modules. No CISA KEV listing and no public PoC are present in the available data, suggesting exploitation has not been observed at scale. EPSS data was not supplied. The combination of a straightforward attack class and an 8.7 CVSS score warrants prompt patching, but the realistic blast radius is constrained to deployments that expose NLTK's XML-parsing paths to external input.

An attacker identifies an application that accepts user-supplied XML data and routes it through an NLTK module for processing. They submit a classic Billion Laughs payload - a document with a small DTD section defining a chain of entities each referencing the previous, starting from a few hundred bytes - against the endpoint. When NLTK's xml.etree.ElementTree parser processes the document, entity expansion consumes memory exponentially, causing the worker process or host to exhaust available RAM and become unresponsive, denying service to legitimate users. No CVE-specific public PoC is required, as the Billion Laughs payload structure is extensively documented and reproducible from first principles.