ARTEX: The FBI's Warning About an AI Hacking Framework That Breaches Networks in Under 3 Hours
An open-source penetration testing tool went from a researcher’s GitHub project to a documented ingredient in nation-state attacks in about four months. The FBI thinks that’s worth a FLASH advisory, and it’s worth paying attention to — regardless of how you feel about anything else going on in Washington right now.
What ARTEX Does
ARTEX is an autonomous pen-testing framework written in Go. Instead of running its own model, it plugs into external LLMs — Claude, GPT, DeepSeek, GLM, Grok, Kimi — and lets AI agents handle the mechanical parts of an intrusion: reconnaissance, vulnerability discovery, exploitation, privilege escalation, data exfiltration. An operator describes a goal; the framework turns it into the sequence of actions that used to take a skilled human operator days to execute by hand.
It was released in July 2026 by a developer known as Autumn-27 and pulled from GitHub in October after reports surfaced that threat actors were using it in the wild. That didn’t stop much — forks, archived mirrors, localized versions, and a commercial “ARTEX PRO” edition are all still circulating. Researchers have already found more than 1,600 publicly exposed ARTEX consoles, and the real number of deployments is believed to be considerably higher.
The FBI’s 13-page advisory, published November 13, 2026, says threat actors have been weaponizing ARTEX since at least August against financial institutions, government agencies, defense contractors, tech companies, telecom and internet providers, and law firms. The headline number: some intrusions went from initial targeting to privileged access inside a network in under three hours.
The Attacks It’s Already Powered
This isn’t theoretical. The advisory ties ARTEX to several already-reported breaches:
- ConnectWise supply-chain attack (October 2026): Microsoft attributed this to Storm-0940, a China-aligned group, which used ARTEX to accelerate reconnaissance and credential abuse after compromising application identities tied to ConnectWise’s cloud backup service (formerly SkyKick). The attackers used those compromised app identities to authenticate against more than 2,250 downstream Microsoft 365 tenants. Huntress identified at least 162 compromised tenants.
- South Korean bank breaches: A Chinese-speaking threat actor combined ARTEX with Claude Code agents to target Shinhan Bank, KB Kookmin Bank, and Hana Bank, exposing personal data on roughly 67,000 people. CrowdStrike found the attacker’s exposed infrastructure — including ARTEX configs and Claude Code session histories — sitting in the open.
- Vapor Panda campaign: CrowdStrike linked this Chinese cyberespionage group to ARTEX-driven attacks against more than 200 organizations across 14 countries, with at least nine confirmed compromises involving reconnaissance, exploitation, web shells, and credential theft.
- South Korean arrests: Police arrested seven people connected to attacks on 90+ organizations — hospitals, schools, businesses — that stole roughly 8.2 million data records, with ARTEX reportedly used in most of the intrusions.
Why the Three-Hour Number Matters
Security teams are built around the assumption that there’s time between initial access and real damage — time to notice unusual traffic, time to isolate a system, time to call someone. Automated multi-target attacks compress that window and run it against dozens of organizations simultaneously, which is a different problem than a single skilled attacker taking their time against one target.
The FBI’s framing is specific: ARTEX doesn’t introduce a new class of vulnerability. It’s exploiting the same overlooked things attackers always have — forgotten partner portals, legacy services, unpatched third-party integrations — just faster, and at a scale that used to require a team.
The Staffing Context
Here’s the part that’s drawing attention beyond the usual security press: the agency built to coordinate the federal response to exactly this kind of threat has shed a documented third of its staff over the past roughly fourteen months. CISA’s authorized workforce fell from around 3,400 at the start of FY2025 to roughly 2,300–2,400 by early 2026, per reporting from the Cloud Security Alliance’s FY2026 budget analysis and congressional testimony — a mix of DOGE-driven cuts, buyouts, and attrition. Cuts reportedly included CISA’s red-team specialists (the people who simulate attacks to find holes before real attackers do) and the agency’s election-security program, whose roughly 14 specialized positions were eliminated heading into the 2026 midterms.
To be clear about what is and isn’t established: the FBI advisory doesn’t connect ARTEX specifically to CISA’s staffing levels, and we’re not aware of reporting that draws a direct causal line between the two. They’re each independently documented. But it’s a fair question to sit with: a surge in AI-automated attacks against exactly the sectors CISA is supposed to help defend is landing at the same time that agency is running at two-thirds strength.
What To Actually Do
Setting aside the politics, the FBI’s recommendations are concrete and worth implementing regardless of your organization’s size:
- Audit application identities and OAuth grants, not just passwords. The ConnectWise breach ran through a compromised app identity with standing permissions — the kind of access that doesn’t show up in a normal login audit.
- Lock down internet-facing partner portals and legacy services. These are ARTEX’s favorite entry points precisely because they’re the systems nobody’s watching closely.
- Rate-limit sensitive APIs and watch for bursts of automated requests rather than individual suspicious logins — the attack pattern here is volume and speed, not stealth.
- Pull the free detection pack. Researcher MalwareMustDie published an open-source ARTEX detection signature pack with IOCs and a confidence-rated detection playbook — a reasonable starting point even if you don’t have a dedicated threat-intel team.
The tool that made this possible is open source and already out of the box it came in. The response to it — at the organizational level, at least — doesn’t have to wait for anyone in Washington to make up their mind about anything.
Sources: BleepingComputer, TechCrunch, Microsoft Security Blog, Cloud Security Alliance FY2026 CISA budget analysis, ProPublica.