AI, Scripts, Operator: Inside the ConnectWise Attack | Huntress

The threat actor Microsoft tracks as Storm-0940 has, according to Huntress, been using ARTEX — an open-source multi-agent AI framework — to abuse compromised ConnectWise application credentials and sign into downstream Microsoft 365 tenants at machine speed. 'At machine speed' is the sort of phrase that turns up in a vendor blog when the incident timeline has already embarrassing gaps in it; the logs arrived, the invoices arrived, and somewhere in between a human being was meant to notice. Administrators are still being told to rotate credentials, which is the security equivalent of being advised to change the locks on a house with no walls.

The genuinely interesting bit, and it is interesting in the way a gas leak is interesting: ARTEX runs LLM-driven agents through a penetration test from reconnaissance to data exfiltration with, in Huntress's words, little needed human interaction. The red team's toolkit has acquired a work ethic and no interest in a pension. Huntress also says its telemetry aligns with Microsoft's findings, which is the closest thing this industry has to peer review — two vendors, one attribution, and a shared reluctance to name whoever was actually holding the credentials.

Note the date: from at least September 2026. 'At least' is doing the heavy lifting there, the way it always does. Open-source frameworks do not have patch windows, support contracts, or anyone whose job depends on them; they have a repository and a fanbase. The result is that a mid-tier criminal crew now runs a penetration test at a tempo that would embarrass most consultancies billing by the day. Nobody had to build a superintelligence for this; they just had to download one. The patch backlog belongs to nobody, which is precisely the design, and a framework that automates a full intrusion chain will be maintained by exactly the sort of people you would least like to stand behind at a conference bar.