
Cloud Vulnerability DB
A community-led vulnerabilities database
CVE-2026-128134 is an untrusted search path vulnerability in NLTK (Natural Language Toolkit) affecting the nltk.parse.dependencygraph.dot2img function. Prior to version 3.10.4, dot2img resolves the Graphviz dot executable through PATH using find_binary but invokes the resulting absolute path via a raw subprocess.run call, bypassing the nltk.pathsec.spawn_trusted trusted-execution chokepoint. This omits ownership/writability validation and loader-environment scrubbing, allowing a local attacker who can write to a directory on the victim's PATH to plant a trojan dot binary that executes as the victim. The vulnerability affects all NLTK versions up to and including 3.10.3 and is fixed in 3.10.4. It carries a CVSS v3.1 base score of 7.8 (High) (GitHub Advisory).
The root cause is classified as CWE-426 (Untrusted Search Path) and CWE-427 (Uncontrolled Search Path Element). NLTK's security model applies four defensive layers to native-tool wrappers: (1) find_binary refuses CWD-relative matches, (2) resolve_trusted_executable verifies file ownership and writability, (3) shell execution is refused, and (4) safe_env scrubs loader environment variables. The dot2img function applies only layer 1 via find_binary("dot"), then passes the resolved absolute path directly to subprocess.run([dot_binary, "-T%s" % t], ...) with no env= argument, skipping layers 2–4. Because layer 1 only rejects non-absolute matches, an absolute path pointing to a file in a world-writable directory passes through unchecked. This is a residual bypass after CVE-2026-78680, which hardened the sibling nltk/translate/api.py invoker through spawn_trusted but left dot2img unprotected. The vulnerability also exposes a loader-injection surface since the child process inherits the full unscrubbed environment, including variables like LD_PRELOAD and DYLD_* (GitHub Advisory, Fix Commit).
Successful exploitation yields arbitrary local code execution as the victim user, with full confidentiality, integrity, and availability impact (all rated High). An attacker can execute any command in the victim's security context, enabling credential theft, data exfiltration, persistence mechanisms, or lateral movement on shared POSIX systems. The attack is particularly dangerous in multi-user environments such as shared research servers or Jupyter notebook deployments, where DependencyGraph._repr_svg_ auto-invokes dot2img on graph display, requiring no explicit victim action beyond rendering a dependency graph. The unscrubbed child environment additionally provides a loader-injection surface via LD_PRELOAD, DYLD_INSERT_LIBRARIES, and similar variables (GitHub Advisory).
A proof-of-concept exploit is publicly available in the GitHub Security Advisory, consisting of a runnable Python script that plants a trojan dot shell script in a world-writable temp directory, prepends it to PATH, and demonstrates execution via dot2img by creating a marker file. The PoC confirms that resolve_trusted_executable correctly refuses the planted binary (showing the hardened path would block it), while the vulnerable dot2img executes it. The EPSS score is 0.0, and there is no evidence of in-the-wild exploitation or CISA KEV catalog listing as of the advisory date. Exploitation requires local access and the ability to write to a directory on the victim's PATH, making it a local privilege escalation or lateral movement primitive rather than a remote attack (GitHub Advisory).
/usr/local/bin with drwxrwxr-x permissions, or any world-writable directory prepended to PATH).dot script: Write a malicious shell script named dot into the writable PATH directory. The script should perform the attacker's desired action (e.g., create a reverse shell, exfiltrate data, or drop a marker file) and optionally emit fake SVG output to avoid detection:#!/bin/sh
touch /tmp/MARKER # or any malicious command
echo '<svg><!-- planted dot ran --></svg>'chmod 0o777 /path/to/writable_dir/dot to ensure the script is executable.export PATH=/writable/dir:$PATH. On a shared system, this may already be satisfied by existing PATH configuration.dot2img: The victim triggers exploitation by rendering a DependencyGraph object in a Jupyter notebook (which auto-invokes _repr_svg_ → dot2img) or by calling dot2img() directly. No control over NLTK inputs is required.dot script executes as the victim user. Confirm via the marker file or observe the attacker-controlled output returned by dot2img (GitHub Advisory).dot in world- or group-writable directories on the system PATH (e.g., /tmp/*, /usr/local/bin/, or custom PATH entries); marker files or unexpected artifacts created in /tmp or the attacker's staging directory; new scripts or binaries with 0o777 permissions in PATH directories./bin/sh, bash, curl, wget, nc) where the parent is a Jupyter kernel or Python interpreter; processes named dot executing from non-standard locations (outside /usr/bin/, /usr/local/bin/ or the standard Graphviz install path).dot binary from a world-writable or temp directory; Python process logs showing dot2img calls followed by unexpected subprocess activity; Jupyter server logs showing _repr_svg_ invocations coinciding with anomalous child process spawns.DependencyGraph rendering operations (GitHub Advisory).Upgrade NLTK to version 3.10.4 or later, which routes dot2img through nltk.pathsec.spawn_trusted (commit b002a72), applying ownership/writability validation and loader-environment scrubbing identical to the hardened sibling in nltk/translate/api.py. Install via pip install --upgrade nltk. As an interim workaround without a code change, ensure no group- or world-writable directory appears on the PATH of any account that renders NLTK dependency graphs, and install Graphviz only in directories writable exclusively by root or the owning user. Administrators should audit PATH directories for unexpected write permissions and educate users against rendering untrusted DependencyGraph objects in interactive environments like Jupyter notebooks (GitHub Advisory, NLTK v3.10.4 Release, Fix Commit).
The vulnerability was reported by security researcher @manus-pi and disclosed via GitHub Security Advisory on October 26, 2026. The fix was merged via PR #3887 as part of a broader security hardening effort in NLTK 3.10.4, which addressed numerous security issues across the library. Social media activity was observed on Mastodon and Bluesky shortly after NVD publication on November 13, 2026, with automated CVE tracking accounts noting the advisory. No significant independent researcher commentary or media coverage beyond the official advisory has been identified (GitHub Advisory, NLTK v3.10.4 Release).
Fix availability across major Linux distributions and their releases.
Source: This report was generated using AI
Free Vulnerability Assessment
Evaluate your cloud security practices across 9 security domains to benchmark your risk level and identify gaps in your defenses.
Get a personalized demo
"Best User Experience I have ever seen, provides full visibility to cloud workloads."
"Wiz provides a single pane of glass to see what is going on in our cloud environments."
"We know that if Wiz identifies something as critical, it actually is."