NadMesh Botnet Targets Exposed AI Services for Cloud Credentials and Kubernetes Tokens

A new Go-based botnet named NadMesh has been discovered actively hunting exposed AI services to steal cloud credentials and Kubernetes tokens. First observed in early July 2026, the botnet’s operator dashboard claims 3,811 unique AWS keys harvested. The malware, named after the ‘n4d mesh controller’ string in its source code, was analyzed by QiAnXin’s XLab, which published a report detailing its capabilities.

NadMesh uses a Shodan harvester to scan for exposed AI services including ComfyUI, Ollama, n8n, Open WebUI, Langflow, and Gradio. The botnet targets cloud keys from environment variables, Kubernetes service account tokens, and configuration files such as ~/.aws/config, .env, and ~/.docker/config.json. The operator’s priority order for exploitation is MCP (Model Context Protocol), followed by Kubernetes, Docker API, and Redis.

XLab observed that the majority of exploit traffic targets Docker containers API RCE (30.31%) and Jenkins script text RCE (22.28%), with Telnet weak passwords (10.36%) and Redis (8.29%) also significant. The botnet employs multiple persistence mechanisms, Garble obfuscation, UPX packing, and random padding to evade detection. Indicators of compromise include C2 at 209.99.186[.]235, domain cdnorigin[.]net, and agent sample SHA1 31c69b3e12936abca770d430066f379ec1d997ec.

Organizations are advised to secure exposed services behind authentication, patch vulnerabilities including CVE-2026-39987 and CVE-2026-41176, and monitor for unauthorized access to cloud credentials and Kubernetes tokens.