The FBI's November 13 FLASH alert says criminals and state-linked groups have used ARTEX, an open-source AI agent framework that runs on models like Claude, GPT and DeepSeek, since at least August 2026. It names South Korean banks and ConnectWise Cloud Backup customers as victims and tells companies to inventory partner portals, legacy apps and admin panels first.
Inventory every internet-facing app your company runs, including the old ones nobody owns. That is the first step in a FLASH alert the FBI published on November 13 about ARTEX, an AI agent framework that attackers use to break into companies with little human effort.
The alert, FLASH-20261113-001, says a range of criminal and state-linked groups have used ARTEX since at least August 2026. Targets span finance, government, tech, healthcare, manufacturing, telecom and legal firms. Some intrusions went from first scan to privileged access in under three hours, according to the FBI. BleepingComputer reported the alert on November 15.
What ARTEX is
ARTEX is an open-source penetration testing tool written in Go. An operator types a goal in plain language and the framework hands it to AI agents that scan, exploit, escalate and pull data. It has no model of its own. The FBI says it plugs into Anthropic’s Claude, OpenAI’s GPT, DeepSeek, GLM, Grok and Kimi, and that attacker-run copies mostly use DeepSeek’s low-cost Flash models.
A developer known as “Autumn-27” released it on GitHub in July 2026 for authorized testing. The developer pulled the repository on October 8 after reports of abuse. Forks, archived copies, localized versions and a paid edition called ARTEX PRO are still available. Researchers counted more than 1,600 ARTEX consoles exposed on the internet in early November, and the FBI says the real number of installs is likely much higher.
Who got hit
The FBI names two campaigns:
- South Korean banks. A financially motivated group exfiltrated personal data on about 67,000 people. BleepingComputer, citing CrowdStrike, names Shinhan Bank, KB Kookmin Bank and Hana Bank as targets.
- ConnectWise Cloud Backup customers. A supply-chain attack on the backup service, formerly SkyKick, gave access to more than 2,250 downstream Microsoft 365 tenants, according to the FBI’s summary of Microsoft’s research.
The alert also lists IP addresses, a domain and behavior patterns from these campaigns. The FBI warns that the IP addresses may already belong to someone else, so vet them before you block them.
Small teams get attacked through the apps nobody watches
This is our analysis. ARTEX does not use new exploits. The FBI’s table of observed activity is a list of old mistakes: unsigned SAML assertions, XXE and SSRF flaws, anonymous file uploads, hardcoded admin credentials and over-permissioned cloud keys. What changed is speed and cost. A small group can now run that checklist against many companies at once, around the clock.
The FBI says ARTEX does best against “peripheral systems”: partner portals, mobile backends, employee apps, sales tools, legacy services and third-party integrations. At a 10- to 50-person company those are the admin panel a contractor set up, the staging site still on a public IP and the OAuth app someone approved two years ago.
What to do this week
The FBI’s recommendations, trimmed to what a small team can act on:
- List everything that faces the internet. Include staging sites, old marketing tools, partner portals and anything a past contractor built.
- Move admin panels off the open internet. Put cloud consoles, backup platforms, deployment endpoints and device management behind a VPN or a zero-trust access tool.
- Audit app permissions in Microsoft 365 or Google Workspace. Remove third-party apps you no longer use and cut broad consent grants.
- If you used ConnectWise Cloud Backup or SkyKick, check every tenant. The FBI says to search by application ID, not display name, even if you stopped using the service.
- Watch for unexpected AI API traffic. The alert lists outbound calls to api.anthropic.com, api.deepseek.com, open.bigmodel.cn and api.x.ai as worth investigating when they come from servers that have no reason to call them.
Report suspected activity to your local FBI field office or at ic3.gov. The full alert is on IC3: FLASH-20261113-001 (PDF).
Stolen AI keys are a related risk. Railway and Northflank both reported breaches in the past three weeks in which attackers used customers’ AI API keys. The VOLTA infostealer uses an AI model to decide which victims are worth the most, and the FBI warned this week about North Korea’s fake Zoom and Teams calls.
