THE CRUNCH

Barracuda has disclosed an intrusion carried out with Artex, an agentic AI penetration testing framework that Chinese threat actors have adopted for real-world attacks. The attackers used an exposed valid cloud access key to map Barracuda's reach, then reused discovered secrets to reach more systems and steal further credentials. In all, the attacker touched about 10 cloud accounts in roughly 44 hours between 1 and 2 November, and Barracuda says the work matched Artex's built-in routines.

Artex, published on GitHub in July 2026, uses a planner to set an objective and create tasks, with agents carrying them out while the platform records progress and findings. It reportedly launched with DeepSeek v4.1-flash but can run other models, and users mainly rely on Chinese models, which are more permissive than US-made ones. Although built for security testing, it has appeared in live intrusions since at least August, and its use has increased significantly in recent weeks.

The FBI has issued an alert warning that the framework lets actors with minimal technical expertise run complex, multi-target operations simultaneously, accelerating attack velocity and expanding scale. It cites attacks on South Korean financial firms and ConnectWise, and flags faster automation, exploitation of neglected peripheral devices and supply chain attack vectors. The developer removed the GitHub repository in October, but Artex continues to spread through other channels; in early November Censys counted roughly 1,500 hosts running exposed Artex consoles.

SentinelOne reports that in October the Chinese espionage group Vapor Panda used Artex in a campaign targeting over 200 organisations in 14 countries, including foreign ministries, a European presidency, and US pharmaceutical, semiconductor, defence and nuclear organisations. Cisco's Talos notes that financially motivated groups such as Storm-3358 have adopted it too. Post-exploitation, attackers abused compromised credentials, escalated privileges, deployed persistence and moved laterally, with impacts ranging from access to cloud secrets and mailbox data to reaching connected environments.

At Barracuda, the attackers targeted secret stores, configurations, Kubernetes service accounts and server metadata services, left a backdoor in a Kubernetes cluster, and on some servers searched for Ethereum keystore files, seed phrases and other wallet information. Barracuda has not attributed the attack, noting that many groups use Artex, and says the intrusion resembles the October breach at Railway and fits a previously reported wave of intrusions at dozens of AI developer platforms.