Chainstack says its intruder used ARTEX, an AI agent system that automates intrusions

Chainstack, a company that provides infrastructure for blockchain networks, has expanded its report on the intrusion it suffered in late October with technical details of the attack. According to the company, the attacker used ARTEX, an open-source framework that coordinates agents based on language models to automate intrusions, and its "artx" and "artex" naming markers appear in the artifacts left in its systems.

According to Chainstack's account, the attacker got in on October 26 through an internet-exposed test Grafana instance that still had its default administrator credentials. From there it reached an internal MCP tool server for AI agents that required no authentication and had cluster administrator rights. Using credentials obtained in the development environment, it reached production on October 29. The company says it found no evidence that any software vulnerability was exploited, but rather a chain of default credentials, exposed functions, missing authentication and excessive permissions.

The intrusion was detected on October 31, when a database replication pipeline failed, and access was cut the same day, according to the company. Chainstack says the attacker accessed a database with organization and account holder names, members' email addresses and password hashes, and had access to systems holding node credentials, although it has no evidence they were read.

The company says customer services operated normally during the incident and is asking all customers to change their passwords, rotate node credentials and regenerate API keys. Chainstack places the incident within a broader campaign of AI-powered attacks reported in late October; The Hacker News links it to other cases attributed to ARTEX, such as WorkOS.