Network-AI AgentVCR Path Traversal Vulnerability Leading to Arbitrary File Read/Write (CVE-2026-120312)
Executive & SecOps Summary
SECOPS ASSESSMENTNetwork-AI versions from 5.13.4 before 5.15.5
Attackers can manipulate the cassette name to read or overwrite arbitrary .json files, potentially altering AuthGuardian trust-level configurations to escalate privileges
Immediately upgrade Network-AI to version 5.15.5 or higher, and implement strict input validation on cassette names
Affected Products & Version Scope
Remediation & Live Workaround Commands
PATCH & WORKAROUND1. Official Patch & Version Upgrade Commands
The recommended remediation is to upgrade the Network-AI package to version 5.15.5 or later.
# Upgrade Network-AI using pip
pip install --upgrade 'network-ai>=5.15.5'
# Or update your requirements.txt/dependency lockfile to enforce the secure version
# network-ai>=5.15.52. Temporary Workaround & Hardening Configuration
If an immediate upgrade is not feasible, implement strict input validation to sanitize the cassette name parameter before passing it to AgentVCR APIs.
import os
def secure_load_cassette(cassette_name):
# Extract only the base filename to prevent directory traversal
safe_name = os.path.basename(cassette_name)
if safe_name != cassette_name:
raise ValueError('Invalid characters detected in cassette name.')
# Ensure the resolved path remains within the designated directory
base_dir = os.path.abspath('./cassettes')
target_path = os.path.abspath(os.path.join(base_dir, safe_name))
if not target_path.startswith(base_dir):
raise PermissionError('Access denied: Path traversal attempt detected.')
return target_path3. Log Audit & IoC Verification Commands
Use the following commands to verify the installed version of Network-AI and inspect application logs for potential exploitation attempts.
# 1. Check the currently installed version of Network-AI
python3 -c "
try:
import network_ai
print('Installed Version:', getattr(network_ai, '__version__', 'Unknown'))
except ImportError:
print('Network-AI is not installed.')
"
# 2. Scan application logs for path traversal patterns (e.g., ../, ..\, ..%2f)
grep -rnE '\.\./|\.\.\\|\.\.%2[fF]' /var/log/network-ai/Scan Your Repositories for Transitive Vulnerable Dependencies in 30 Seconds
Automatically detect vulnerable packages across npm, PyPI, Maven, and Go, and open automated fix Pull Requests without breaking your build.
Technical Root Cause & Exploitation Mechanics
1. Vulnerability Overview & Root Cause
This vulnerability (CVE-2026-120312) is a Path Traversal (CWE-23) flaw located within the AgentVCR component of the Network-AI framework. AgentVCR records and replays network interactions using 'cassette' files. The root cause is the lack of proper sanitization and validation on the cassette name parameter before it is resolved into a file system path, allowing directory traversal sequences (e.g., ../) to be processed directly.
2. Exploitation Mechanics & Attack Scenario
An authenticated attacker with low privileges can supply malicious payloads containing relative path traversal sequences to the load() or save() methods of AgentVCR.
- Arbitrary File Read: By passing traversal sequences to
load(), the attacker can read sensitive.jsonconfiguration files or environment variables stored within the application's accessible directories. - Arbitrary File Write & Privilege Escalation: By passing traversal sequences to
save(), the attacker can overwrite critical system files, such as the AuthGuardian trust-level configuration. This allows the attacker to bypass authentication controls and escalate their privileges to administrative levels.
3. Infrastructure & Business Impact
With a CVSS score of 8.8 (High), this vulnerability poses a severe threat to infrastructure security. Successful exploitation can lead to complete compromise of the AI pipeline and connected systems. Overwriting authorization policies can grant unauthorized access to proprietary AI models, sensitive customer data, and internal APIs, resulting in severe compliance violations and operational disruption.
Official Vendor Advisories & References
Network-AI from 5.13.4 before 5.15.5 contains a path traversal vulnerability in AgentVCR that allows attackers controlling the cassette name to read or write arbitrary .json files. Attackers can pass traversal sequences to load() or save() to disclose sensitive JSON data or overwrite policy files such as AuthGuardian trust-level configuration.
Frequently Asked SecOps Questions (CVE-2026-120312)
FAQQ1. How critical is CVE-2026-120312 and is it actively exploited in the wild?
CVE-2026-120312 is rated CVSS 8.8 (HIGH) with a 30-day EPSS exploitation probability of 0.4%. Organizations running affected versions should prioritize patching based on network exposure.
Q2. Which vendors and products are affected by CVE-2026-120312?
Primary impacted vendors include jovancoding across network-ai. Scope summary: Network-AI versions 5.13.4 and prior to 5.15.5
Q3. What temporary mitigations or workarounds exist for CVE-2026-120312 before patching?
[Mitigation] Immediately upgrade Network-AI to version 5.15.5 or higher, and implement strict input validation on cassette names Follow the CLI remediation commands and interactive SecOps checklist on this page to restrict exposure prior to your maintenance window.
Generate an Internal Engineering Notice or CISO Report for CVE-2026-120312
Instantly format this advisory into an executive CISO summary or engineering patch ticket.
🔗 Related Advisories in Similar Stack (jovancoding)
View Category →Network-AI FileJobStore Path Traversal Vulnerability Analysis and Mitigation Guide (CVE-2026-120311)
[Affected Scope] Network-AI versions from 5.13.4 before 5.15.5
Network-AI InputSanitizer Path Traversal Vulnerability (CVE-2026-120313) Technical Analysis and Mitigation Guide
[Affected Scope] Network-AI versions prior to 5.15.5
sevenz-rust2 Path Traversal and Arbitrary File Overwrite Vulnerability (CVE-2026-61725)
[Affected Scope] Rust applications utilizing sevenz-rust2 library versions prior to 0.21.1.
GitLab CE/EE Unauthenticated Remote Code Execution (RCE) Advisory (CVE-2026-97229)
[Affected Scope] GitLab CE/EE versions 19.2 (< 19.2.8), 19.3 (< 19.3.5), and 19.4 (< 19.4.2)
Subscribe to Daily Vulnerability Briefings & KEV Alerts
Concise morning briefings for SecOps, DevSecOps, and SRE teams covering newly exploited CVEs and mitigation priorities.