The U.S. Federal Bureau of Investigation (FBI) has published a FLASH alert (FLASH-20261113-001), dated November 13, 2026 (U.S. time), on ARTEX, an autonomous penetration testing framework built on AI agents. The FBI says a tool meant for security testing has been turned against organizations, automating everything from reconnaissance to data exfiltration at machine speed, and it is urging organizations to take defensive steps.

The FBI Has Tracked Abuse Since August 2026 Across More Than 10 Sectors

According to the FBI, a variety of malicious actors have used ARTEX since at least August 2026 to target organizations around the world. The sectors named are financial services, government, technology, defense, semiconductor, energy, healthcare, manufacturing, telecommunications and legal.

The FBI says ARTEX sharply cuts the time needed to chain attack techniques together. Some intrusions went from initial targeting to privileged access in under three hours. The agency adds that the framework lets actors with minimal technical expertise run complex operations against multiple targets at the same time.

The FBI highlighted two cases. One was a financially motivated campaign against South Korean financial institutions in which personal information on about 67,000 people was exfiltrated. The other was a supply chain compromise affecting customers of ConnectWise Cloud Backup that enabled access to more than 2,250 downstream tenants.

ARTEX Turns Natural-Language Goals Into Executable Commands

As the FBI describes it, ARTEX is an autonomous, multi-agent penetration testing platform written in Go. A user states high-level goals in natural language, and specialized agents translate them into concrete commands and run them. Beyond setting the initial objective, little human involvement is needed.

ARTEX has no AI model of its own; it calls out to external large language models (LLMs). The FBI lists Anthropic's Claude and OpenAI's GPT, along with DeepSeek, GLM, Grok and Kimi. Analysis of attacker-controlled ARTEX instances found Chinese models to be the most common, with DeepSeek's Flash family of models standing out. The reasons given are their balance of capability and cost, and the lack of visible guardrails when requests are framed as penetration testing.

ARTEX was released on GitHub in July 2026 by a developer known as "Autumn-27." The developer said the tool was intended for authorized security testing within defined scopes, but took the repository down on October 8 after reports of abuse. Even so, it remains in circulation through community forks, localized variants, archived copies and a commercial edition called "ARTEX PRO." The FBI says that as of early November 2026, more than 1,600 publicly exposed ARTEX consoles had been found, and that total deployments, including those behind firewalls, are likely much higher.

Microsoft Found ARTEX Was Used in the ConnectWise Attack

On the ConnectWise case, Microsoft published an analysis on its official security blog on October 21. Microsoft assessed that Storm-0940, an actor aligned with the Chinese state, was behind the supply chain compromise. In Microsoft's analysis, Storm-0940 used ARTEX to speed up scanning and reconnaissance, password spraying and OAuth application abuse at downstream victims, and used already-authorized applications to attempt sign-ins to more than 2,250 tenants.

Overlooked Peripheral Systems Are the Most Exposed

The FBI says ARTEX is especially effective against peripheral systems that get less security attention than flagship applications. Examples include partner and supplier portals, mobile backends, employee applications, sales-support systems, legacy services, third-party integrations and applications built by different business units. The automated traffic is noisy and can set off alerts, but the speed of the attacks leaves defenders little time to coordinate a response.

Key steps recommended by the FBI include:

  • Inventory every internet-facing application, including peripheral systems
  • Restrict access to cloud administration, network device management, backup platforms and similar services to trusted networks or dedicated administrative paths
  • Apply least privilege to multi-tenant application permissions and consent grants, and monitor credential additions and privilege assignments
  • Put rate limits on high-impact APIs and alert on sessions that access unusually large numbers of records
  • Scan configuration files and CI/CD pipelines for embedded secrets, and rotate any exposed credentials promptly
  • In Microsoft 365 tenants where ConnectWise applications were installed, review activity from at least October 1, 2026, and preserve evidence

The FBI asks anyone who spots suspicious activity to contact their local FBI field office or the Internet Crime Complaint Center (IC3).

Our Take

We see ARTEX as a case of an AI agent released as a legitimate testing tool spreading as an attack tool largely unchanged. Copies and a commercial edition survive the original repository's removal, so stopping distribution does not end the problem. As attacks require less effort and know-how, techniques once aimed mainly at large organizations become easier to turn on smaller companies and their suppliers. Organizations in Japan can also be drawn in through overseas partners and cloud services. Operations that rely on manual monitoring will struggle to keep pace with a three-hour timeline, so basic measures such as inventorying exposed assets and reviewing app permissions will likely need to go hand in hand with automated detection.

Summary

The FBI has warned that ARTEX, an AI agent-based penetration testing framework, has been used since August 2026 in attacks across a wide range of sectors. Paired with external LLMs, ARTEX automates everything from reconnaissance to exfiltration, and some intrusions have reached privileged access in under three hours. The public repository is gone, but copies are still circulating, and organizations are being urged to inventory exposed systems, minimize permissions and monitor their APIs.