Attackers Hijack Cairn AI Pentesting Agents to Deploy Backdoors and Steal API Keys

Attackers hijacked exposed Cairn AI-agent management panels, submitting malicious tasks that abused legitimate command-execution and file-access capabilities to install unauthorized SSH keys, deploy malware, and steal credentials, including AI model API keys. NorForge observed malicious tasks on at least 17 exposed panels, including red-team and capture-the-flag environments, with activity beginning October 20 and possible earlier probing. The attackers turned agent-management access into a pathway for persistence and credential theft.

NorForge identified two activity clusters: AIA0009, which pursued persistence and was the only cluster observed deploying the Go-based Linux Poseidon remote access trojan, and AIA0010, which searched for credentials, internal services, and additional exposed panels. Synchronized activity suggested automation, but researchers also saw signs of human intervention and did not establish that AI operated the campaigns. Organizations running these agents should remove public panel exposure, isolate task execution, restrict agent capabilities, rotate potentially compromised API keys, and inspect SSH authorized_keys files for unauthorized entries.