Path Traversal in Network-AI AgentVCR

Executive Summary

This vulnerability is a path traversal issue in Network-AI's AgentVCR component affecting versions 5.13.4 to 5.15.4. It allows attackers who control the cassette name to read or write arbitrary .json files outside the intended directory by using traversal sequences like ../../etc/some-config. The methods load() and save() do not validate paths, enabling unauthorized file access or modification.

Useful 0 Not Useful 0

Detection Guidance

Check if Network-AI versions 5.13.4 to 5.15.4 are installed using package managers like npm or pip. Inspect AgentVCR usage in applications for untrusted cassette name inputs. Monitor filesystem access logs for unexpected .json file reads or writes outside intended directories.

Useful 0 Not Useful 0

Impact Analysis

Attackers can exploit this to read sensitive JSON data such as credentials or write to policy files like AuthGuardian trust-level configurations. This could lead to data breaches, configuration corruption, or token invalidation in multi-agent systems. The impact includes high confidentiality and integrity breaches.

Useful 0 Not Useful 0

Compliance Impact

This vulnerability could lead to unauthorized access to sensitive data, violating GDPR's data protection requirements and HIPAA's confidentiality rules. Exposure of personal or health data may result in compliance breaches, legal penalties, and reputational damage.

Useful 0 Not Useful 0

Mitigation Strategies

Upgrade Network-AI to version 5.15.5 or later immediately. Implement strict input validation for cassette names, rejecting any containing path traversal sequences like ../. Add filesystem access controls to restrict AgentVCR to only permitted directories.

Useful 0 Not Useful 0