GL-iNet GL-MT3000 s2s.so Command Injection (CVE-2026-18614)

Technical Analysis

The vulnerability resides in the /cgi-bin/glc CGI endpoint, where the s2s.so native plugin fails to sanitize the 'port' parameter passed to enable_echo_server, allowing shell metacharacters or command sequences to be executed with the privileges of the CGI process. Because the attack surface is a network-facing CGI interface, no authentication or user interaction appears to be required, making mass scanning and automated exploitation highly feasible. Public availability of a working exploit significantly raises the likelihood of opportunistic and botnet-driven attacks (e.g., Mirai-style IoT worms) against internet-exposed devices. Organizations that deploy GL-MT3000 routers as edge or travel connectivity devices for remote workers, branch offices, or IoT/edge AI deployments risk device takeover, traffic interception, and lateral movement into internal networks. If AI agent systems or LLM-based automation rely on network connectivity, API access, or remote management through these compromised routers, attackers could intercept or manipulate agent traffic, harvest embedded API keys/credentials, or pivot into agent infrastructure, making this agent-relevant in edge and IoT-connected agent deployments.

Affected Systems

GL-iNet GL-MT3000 routers running firmware up to and including version 4.4.5, specifically the s2s.so native plugin component and its exposed /cgi-bin/glc CGI interface.

Indicators of Compromise

• Endpoint: /cgi-bin/glc
• Parameter: port (command injection vector)
• Component: s2s.so Native Plugin
• Function: s2s.enable_echo_server

Remediation Steps

1. 1

   Apply Vendor Patch

   Update GL-MT3000 firmware to a version beyond 4.4.5 once the vendor releases a patched build addressing the s2s.so command injection flaw.

2. 2

   Restrict Remote Access

   Disable remote/WAN-facing administration of the router's CGI interface and restrict access to trusted internal networks or VPN only.

3. 3

   Network Segmentation

   Isolate router management interfaces from segments hosting AI agent infrastructure, RAG pipelines, or credential stores to limit lateral movement risk.

4. 4

   Monitor for Exploitation

   Inspect device logs and network traffic for anomalous requests to /cgi-bin/glc with unusual 'port' parameter values indicative of command injection attempts.

5. 5

   Credential Rotation

   Rotate any API keys, VPN credentials, or agent-related secrets that may have transited or been stored on potentially compromised routers.