NLTK (Natural Language Toolkit) is a suite of open source Python modules, data sets, and tutorials supporting research and development in Natural Language Processing. Prior to 3.10.4, nltk.parse.dependencygraph.dot2img resolves the Graphviz dot executable through PATH but invokes the resulting absolute executable without routing through spawn_trusted, so resolve_trusted_executable ownership and writability validation and safe_env loader environment scrubbing do not run. A local low-privileged attacker who can write to a directory on a POSIX victim's PATH can plant or replace dot, and code executes as the victim when dot2img is called directly or DependencyGraph.repr_svg renders a graph. The current-working-directory-only bare-name vector is blocked and is not this vulnerability; this issue is the residual writable absolute PATH element bypass after CVE-2026-78680. This issue is fixed in version 3.10.4.
NLTK from 3.10.0 before 3.10.4 contains an arbitrary file read vulnerability in BllipParser that allows attackers to open files outside NLTK data roots by supplying unvalidated model paths. Attackers controlling parser_model, reranker_features, reranker_weights, or from_unified_model_dir() input can make the bllipparser C extension read sensitive files or block on FIFOs and devices.
NLTK before 3.10.4 contains a path traversal vulnerability in save_maxent_params() and load_maxent_params() that bypasses the pathsec filesystem restriction mechanism. Attackers who control the tab_dir argument can read or overwrite model parameter files outside permitted data roots, because the functions call Python's built-in open() directly.
NLTK before 3.10.4 contains a quadratic-time denial of service vulnerability in ConllCorpusReader._get_srl_instances that allows remote attackers to exhaust CPU through nested span rescans. Attackers can supply a crafted CoNLL SRL corpus with many predicates and span columns, placing the head span last, to pin a core for minutes.
NLTK before 3.10.4 contains a denial of service vulnerability in Tree.chomsky_normal_form that allows remote attackers to exhaust CPU through quadratic-time right-factor binarization using list.pop(0). Attackers can supply a wide flat treebank tree parsed by Tree.fromstring to force tens of seconds of single-threaded CPU consumption, blocking the processing thread.
NLTK before 3.10.4 contains a denial of service vulnerability in the CCG lexicon parser fromstring that allows remote attackers to exhaust CPU by supplying wide inputs. Attackers can submit lexicon text with long flat bracket bodies or operator chains, triggering quadratic string re-slicing in matchBrackets and augParseCategory that blocks worker threads.
NLTK from 3.2.5 before 3.10.4 contains an algorithmic complexity vulnerability in TreebankWordTokenizer.span_tokenize that allows remote attackers to cause quadratic CPU consumption through repeated list.pop(0) calls. Attackers can submit long runs of double-quote characters to services span-tokenizing untrusted text, exhausting a CPU core and degrading availability.
NLTK from 3.6 before 3.10.4 contains an algorithmic complexity vulnerability in LegalitySyllableTokenizer.tokenize that allows remote attackers to exhaust CPU by supplying long all-vowel tokens. Attackers can submit a token such as a long run of vowels, causing the unguarded onset reversal loop to perform quadratic work and deny service to syllabification components.
NLTK from 3.9.4 before 3.10.4 contains a symlink following vulnerability in nltk.sem.chat80.sql_query() that bypasses the pathsec data-root sandbox containment check. Attackers who can write into a trusted NLTK data root can plant a symbolic link to read SQLite databases outside the sandbox.
NLTK before 3.10.4 contains an uncontrolled resource consumption vulnerability in SteppingRecursiveDescentParser.parse() that lacks any max_time deadline or step limit. Attackers influencing the grammar can supply an ambiguous self-embedding grammar and short token input to force unbounded top-down search, pinning CPU cores and exhausting worker pools.
NLTK from 3.10.0 before 3.10.4 contains a symlink escape vulnerability in YCOECorpusReader that allows attackers to read files outside the corpus root by planting psd or pos directory symlinks. Attackers controlling a YCOE corpus can make parsed_sents() and tagged_words() return file content from other corpora or temporary files within the NLTK data sandbox.
NLTK versions up to 3.10.3 have a path traversal vulnerability in the WekaClassifier component. This flaw allows attackers who control the model filename or model argument to bypass path containment checks and cause the Weka JVM to write or probe files outside of permitted directories.
NLTK versions up to and including 3.10.3 have a path traversal vulnerability in certain model-artifact APIs. This flaw allows attackers to bypass path security enforcement and perform unauthorized read or write operations on files outside the intended sandbox. The affected components include TransitionParser, AveragedPerceptron, PerceptronTagger, and maxent parameter APIs when pathsec is enabled.
CVE-2026-72818 is a high-severity vulnerability in the nltk library's TweetTokenizer component. It arises from an inefficient regular expression in nltk/tokenize/casual.py that can cause exponential backtracking when processing certain crafted inputs. This leads to excessive CPU consumption and potential denial of service by stalling services that tokenize untrusted social media text. Version 3.10.1 of nltk addresses this issue by bounding the label repetition in the regular expression.
A Server-Side Request Forgery (SSRF) vulnerability exists in nltk/nltk versions 3.9.4 and the current develop branch. The `nltk.pathsec.validate_network_url()` function, intended to prevent SSRF by rejecting internal network addresses, fails to reject IPs in the RFC 6598 shared address space (`100.64.0.0/10`). This occurs because Python's `ipaddress` module does not classify such addresses as `is_private` or `is_global`, and the current guard only checks `is_private` and a few explicit categories. An attacker who can influence a URL passed to NLTK's network-loading helpers can exploit this vulnerability to make a strict-mode application send requests to shared-address-space hosts, potentially exposing non-public infrastructure reachable from the application host. The impact is limited to SSRF-style confidentiality exposure, with no code execution claimed.
CVE-2026-12261 is a medium severity vulnerability in the nltk Python library's downloader component (nltk.downloader) affecting versions up to 3.9.4. The flaw allows cross-package resource and model poisoning because the downloader extracts package archives into shared namespaces without proper isolation and only validates package integrity after extraction. This enables one package to overwrite another's trusted resources, which remain active across interpreter restarts and can impact workflows relying on NLTK APIs.
CVE-2026-12259 is a medium severity vulnerability in nltk version 3.9.4 where the downloader component may write and extract package data before verifying integrity checks such as SHA-256 or MD5. This flaw allows an attacker who can manipulate the package source (e.g., via a compromised mirror or malicious proxy) to substitute malicious package content. As a result, downstream users or applications may trust and use attacker-controlled corpus or model data.
In nltk/nltk versions 3.9.3 and earlier, five Stanford interface classes (StanfordPOSTagger, StanfordNERTagger, StanfordParser, StanfordDependencyParser, and StanfordNeuralDependencyParser) are vulnerable to untrusted JAR code execution. These classes accept user-controllable JAR paths and execute them via the `java()` function, which invokes `subprocess.Popen()` without integrity verification. This vulnerability is identical to CVE-2026-0848, which was fixed for StanfordSegmenter by adding SHA256 verification. However, the fix was not applied to these additional classes, leaving them susceptible to arbitrary code execution when loading untrusted JAR files.
A vulnerability in nltk's WordNet Browser HTTP server (up to version 3.9.3) allows unauthenticated remote shutdown via a specific GET request, causing denial of service. The server listens on all interfaces and lacks authentication for this critical function. This issue impacts service availability by terminating the server process immediately.
A vulnerability in the `filestring()` function of the `nltk.util` module in nltk version 3.9.2 allows arbitrary file read due to improper validation of input paths. The function directly opens files specified by user input without sanitization, enabling attackers to access sensitive system files by providing absolute paths or traversal paths. This vulnerability can be exploited locally or remotely, particularly in scenarios where the function is used in web APIs or other interfaces that accept user-supplied input.