In a concerning development, cyberattackers are exploiting AI tools designed for network security testing to infiltrate systems. A recent investigation highlights that these intrusions target exposed panels on the Cairn platform, which enables AI agents to function autonomously. This breach poses significant challenges for cybersecurity teams tasked with protecting sensitive information.
How Hackers Compromise AI Tools
Beginning on October 20, 2026, attackers initiated automated campaigns to infiltrate unsecured Cairn panels. These panels, once accessed, allow the attackers to inject tasks that install unauthorized SSH keys, deploy remote access trojans, and search for stored API keys. The Cairn platform’s design, which allows users to define goals and monitor AI agent activities, becomes a vulnerability when left unprotected.
One notable incident involved an individual known as AstraRCE, who defaced several Cairn panels. Although this act of vandalism was separate, it underscored the potential for more severe breaches. Researchers at NorForge, a company specializing in AI agent security, identified these malicious activities while monitoring exposed panels.
Details of the Intrusion Campaign
Initial intrusions appeared as a series of 11 unusual tasks on one panel, marked by hexadecimal strings in their titles. These tasks focused on the local Cairn host, with one instructing an agent to add an unknown public key to a user’s authorized_keys file. This activity quickly disappeared from view, but similar tasks were later found on 17 additional panels, some associated with red team operations.
The campaign evolved with the introduction of malware. On October 21, a task labeled ‘Local worker maintenance’ included a base64-encoded script that downloaded and executed a Linux binary, identified as a Poseidon implant. This Go-based trojan, typically managed via the Mythic red team framework, remained active until deliberately shut down by an operator.
Implications and Security Measures
The intrusions extended beyond malware deployment to involve reconnaissance tasks that searched for sensitive information like API keys. NorForge documented the theft of these secrets, emphasizing the risk inherent in exposed AI tools. The company tracks two distinct groups, AIA0009 and AIA0010, involved in these breaches. While AIA0009 focuses on malware deployment, AIA0010 targets internal services and credentials.
AI’s role in facilitating these attacks remains a topic of investigation, with signs of both automated and human-driven activities. NorForge advises securing Cairn panels, using containerized environments for agent tasks, and regularly updating API keys to mitigate risks.
In light of these findings, organizations utilizing AI in cybersecurity should prioritize securing their infrastructure to prevent unauthorized access. Rotating API keys and reviewing SSH key files are critical to maintaining robust defenses against such sophisticated threats.
