An OS command injection vulnerability in the Network-AI agent sandbox permits authenticated attackers to execute arbitrary commands via improper handling of shell metacharacters.

The vulnerability exists in the ShellExecutor component (CWE-78), which incorrectly processes allowed commands through a shell. This allows an authenticated attacker to append malicious commands to otherwise permitted operations, effectively bypassing security policies.

This vulnerability poses a significant risk to the integrity and security of the orchestrator environment. Successful exploitation allows an attacker with low-level privileges to escape the intended sandbox, potentially leading to full system compromise or lateral movement within the network. The CVSS score of 9.9 reflects the high potential for total system impact.

Immediate Action: Update the Network-AI package to version 5.9.1 or later immediately. This version introduces safer execution patterns that avoid shell invocation.

Proactive Monitoring: Review logs for suspicious command executions, particularly those involving shell metacharacters like semicolons, pipes, or command substitution sequences.

Compensating Controls: Avoid using broad wildcard entries in sandbox allowlists, such as node * or npm *, which provide excessive surface area for potential injection.

Public Exploit Available: No (Exploit available: false)

Organizations utilizing Network-AI must prioritize upgrading to version 5.9.1. Following the update, security teams should review and tighten sandbox policies to adhere to the principle of least privilege, ensuring that only strictly necessary commands are permitted for agent execution.