The FBI has issued a FLASH alert on ARTEX, the open-source autonomous penetration-testing framework, saying a variety of malicious actors have used it since at least August against organizations in financial services, government, defense, semiconductors, energy, healthcare and other sectors. Some of those intrusions went from initial targeting to privileged access in under three hours, the bureau said in the November 13 alert.

ARTEX turns goals written in natural language into commands carried out by AI agents, with little human involvement after the objective is set. It has no model of its own and connects to outside providers, including Anthropic’s Claude, OpenAI’s GPT, DeepSeek, GLM, Grok and Kimi. The FBI said Chinese models are the most common on attacker-controlled instances, “with DeepSeek Flash family models dominating usage due to their balance of capabilities, relatively low cost, and lack of observable guardrails when framed for penetration testing.”

The FBI said financially motivated attacks on South Korean financial institutions led to the theft of personal information on about 67,000 people. It also said a supply chain compromise affecting ConnectWise Cloud Backup customers enabled access to more than 2,250 downstream tenants, citing an October 21 Microsoft report. That report says Storm-0940, which Microsoft describes as a China-state aligned actor, used ARTEX to attempt sign-ins to more than 2,250 tenants after the compromise. Microsoft said most of the group’s ARTEX-driven targeting was unsuccessful but that it gained privileged cloud persistence at many organizations.

A developer using the name “Autumn-27” released ARTEX on GitHub in July and made the repository unavailable on October 8 after reports of abuse. The FBI said the framework still circulates through community forks, localized versions, archived copies and a commercial version, ARTEX PRO. A researcher’s ARTEX agent breached an internal WorkOS tool in October, and Railway, citing Wiz, has said it was defending against a harness similar to ARTEX.

Amazon published its own findings the same day. From October 3 to November 5, its MadPot decoy sensors, set up to look like popular LLM inference APIs, received requests from ARTEX agents. The requests carried the agents’ own records of their campaigns, including system prompts, tool histories and reported findings.

Amazon said it assesses with moderate confidence that the activity came from “disparate Chinese-nexus threat groups,” citing simplified Chinese prompts, China time-zone settings and targeting consistent with Chinese interests. Some instructions told agents to exclude mainland China targets and government domains.

In the most prominent campaign Amazon described, the agents’ records show stolen keys for AI models being fed into the framework’s own model gateway, where they paid for the inference that powered the search for more keys. Targets included LLM gateways and AI application builders such as LiteLLM, Dify, Langflow and Ollama. A second campaign tried to pass identity checks at two cryptocurrency exchanges with real-time deepfake video and altered ID photos. One verification vendor’s server rejected the deepfake in every attempt the agents recorded, according to Amazon, which said it did not observe the agents’ activity as it happened and cannot verify the outcomes they reported.

The FBI tells organizations with the affected ConnectWise applications to keep those connections disabled until ConnectWise issues formal restoration instructions. It asks organizations reporting ARTEX incidents to a local FBI field office or the Internet Crime Complaint Center to include console artifacts, outbound calls to LLM APIs, unfamiliar model proxies or resellers in egress logs, and billing anomalies.

Sources: FBI, Microsoft, Amazon

–
By the Control Plane Editorial Team