CVE-2026-79675 — CVSS 9.8 CRITICAL Severity | NVD Vulnerability Record

Executive Summary

The NLTK Python library (versions < 3.10.3) fails to validate JVM options supplied to its java() function, enabling injection of malicious flags such as -agentpath, -javaagent, or @argfile. This flaw permits unauthenticated remote code execution on any host running vulnerable NLTK‑enabled applications, directly affecting data‑processing pipelines, AI/ML services, and research environments that rely on NLTK. Immediate board‑level decision is required to approve emergency patch deployment or temporary mitigation (e.g., disabling java() usage) across all affected assets.

Verified Facts

• NLTK before version 3.10.3 does not validate JVM options passed via the java() function — NVD record.
• Attackers can inject dangerous JVM flags (e.g., -agentpath, -javaagent, @argfile) to achieve arbitrary code execution — NVD record.
• The vulnerability is assigned CVSS 3.1 base score 9.8 (Critical) with vector AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:H/A:H — NVD record.

Threat Classification

Technical vulnerability (Remote Code Execution) affecting Python‑based data‑science and AI/ML environments; sectors include finance, healthcare, academia, and cloud service providers globally; exploitation status is currently theoretical with proof‑of‑concept code publicly disclosed (HIGH CONFIDENCE); attacker motivation is presumed to be system compromise and data theft (MEDIUM CONFIDENCE).

Threat Severity Assessment

• Exploitability: Network‑remote, no authentication required, low complexity — HIGH CONFIDENCE.
• Impact Scope: Full confidentiality, integrity, and availability compromise (C:H/I:H/A:H) — HIGH CONFIDENCE.
• Prevalence: NLTK is a top‑10 Python library with millions of installations — MEDIUM CONFIDENCE.
• CVSS Rating: 9.8 (Critical) — HIGH CONFIDENCE.

Business Impact

Successful exploitation can lead to full system takeover of servers running vulnerable NLTK, causing immediate disruption of data pipelines, loss of proprietary models, and potential exfiltration of personal data subject to GDPR or NIS2, exposing the organization to fines up to €20 million or 4 % of annual turnover. Compromise of research environments may breach intellectual property agreements, triggering contractual penalties and reputational damage that can erode client trust within weeks.

Technical Analysis

• Attack Vector: Network‑remote injection of malicious JVM flags via the java() API call — directly stated.
• Exploitation Chain: Attacker supplies crafted options → NLTK’s java() forwards unchecked flags to the JVM → JVM loads attacker‑controlled agent or argument file → arbitrary code runs with the privileges of the hosting process.
• Affected Component: NLTK Python package, java() function, versions < 3.10.3.
• Root Cause: Missing validation of per‑call options parameter (CWE‑88: Argument Injection / Improper Neutralization of Argument Injection).
• Impact: Arbitrary code execution (remote) with no authentication required (PR:N) and no user interaction (UI:N).

CVE Analysis

• CVE‑2026‑79675 — NLTK < 3.10.3 — CWE‑88 (Argument Injection) — Network‑remote (AV:N) — No authentication required (PR:N) — Patch introduced in NLTK 3.10.3 (available via PyPI).

MITRE ATT&CK Mapping

• Execution → T1059.007: Java — attacker injects malicious JVM flags to launch arbitrary code via the Java interpreter.
• Privilege Escalation → T1068: Exploitation for Privilege Escalation — arbitrary code runs with the privileges of the compromised Python process.

IOC Intelligence

No public IOCs confirmed at time of publication.

Behavioral IOC categories for rule creation:

• Process creation events where java is invoked with -agentpath, -javaagent, or @argfile arguments.
• Python processes (e.g., python.exe, python3) spawning child processes named java with suspicious command‑line length (>200 characters).
• Audit logs showing NLTK library calls to java() with non‑standard option strings.
• File creation or modification of temporary argument files (e.g., /tmp/@argfile*) immediately preceding a Java process launch.

Detection Engineering Guidance

Log Sources: Windows Sysmon (Event ID 1), Linux auditd (execve), Cloud‑provider function logs, Python application logs.

Detection Logic: Correlate a parent process of type python with a child process java where the command line contains any of the flags -agentpath, -javaagent, or a filename prefixed with @. Flag events where the originating Python script imports nltk and calls java() (if application logging is available).

Sigma Rules

title: NLTK java() Function Abuse – Malicious JVM Flags
id: 9f8c2d3e-4b1a-4f9e-9c2d-6a7b5e1f2c3d
status: stable
description: Detects execution of Java with JVM flags commonly abused via NLTK's java() function.
author: CYBERDUDEBIVASH Sentinel APEX
logsource:
    product: windows
    service: sysmon
detection:
    selection_parent:
        ParentImage|endswith: '\python.exe'
    selection_java:
        Image|endswith: '\java.exe'
        CommandLine|contains:
            - '-agentpath'
            - '-javaagent'
            - '@'
    condition: selection_parent and selection_java
falsepositives:
    - Legitimate data‑science pipelines that intentionally use custom JVM agents
level: high
tags:
    - attack.execution
    - attack.t1059.007

Threat Hunting Queries

• Hypothesis: Malicious JVM flags are being passed from Python to Java — Source: Sysmon Event ID 1, fields ParentImage, Image, CommandLine.
• Hypothesis: Temporary @argfile artifacts are created before Java execution — Source: Windows Sysmon Event ID 11 (FileCreate), filter TargetFilename contains “@argfile”.
• Hypothesis: NLTK version < 3.10.3 is present on production hosts — Source: Endpoint inventory logs, filter PackageName = “nltk” and Version < “3.10.3”.
• Hypothesis: Java processes launched with >200‑character command lines from Python — Source: Linux auditd execve records, fields exe, cmdline, pid.
• Hypothesis: Unexpected network connections from Java processes spawned by Python — Source: Zeek/Suricata logs, filter orig_h where parent_process contains “python” and process_name = “java”.

SOC Analyst Playbook

• P0 – Within 1 hour: Query Sysmon for Event ID 1 where ParentImage ends with “python.exe” and child Image ends with “java.exe” with flagged arguments; isolate affected host.
• P1 – Within 4 hours: Pull endpoint inventory to confirm NLTK version; if < 3.10.3, initiate emergency patch rollout or disable java() calls.
• P2 – Same day: Review file system for creation of “@argfile” temporary files; collect associated process trees for forensic analysis.

Executive Decision Matrix

🔗 Related Intelligence Reports

• Related topic: CVE-2026-78570 — CVSS 9.8 CRITICAL Severity | NVD Vulnerability Record
• Related topic: CVE-2026-79657 — CVSS 9.8 CRITICAL Severity | NVD Vulnerability Record
• Related topic: CVE-2026-63586 — CVSS 9.8 CRITICAL Severity | NVD Vulnerability Record
• Related topic: CVE-2026-78568 — CVSS 9.8 CRITICAL Severity | NVD Vulnerability Record
• Related topic: CVE-2026-78477 — CVSS 9.8 CRITICAL Severity | NVD Vulnerability Record