ARTEX AI Agents Turned Into Cyber Weapons as China-Linked Hackers Target 2,250 Microsoft 365 Tenants
ARTEX, the open-source AI agent framework that helped hackers target thousands of Microsoft 365 accounts through a single ConnectWise breach
Thank you for reading this post, don't forget to subscribe!October 31, 2026|CoreSignal
Hackers linked to China have used an AI toolkit called ARTEX to go after more than 2,250 Microsoft 365 accounts belonging to companies around the world. Microsoft shared the details on October 21 and named the group behind it Storm-0940. The attack shows how AI agents can now do much of a hacker’s work on their own.
How One Backup Tool Became the Way In
The attack started with ConnectWise Cloud Backup, a service that IT companies use to back up their customers’ Microsoft 365 data. The hackers got hold of the login details for ConnectWise’s cloud apps. Those apps already had permission to enter customer accounts, so the attackers simply logged in as them.
Because no real person had to sign in, many normal security alerts never went off. The Register reported that the login attempts hit up to 2,900 a minute, and in some cases the hackers reached admin-level access in under three hours.
What ARTEX Actually Does
ARTEX is a free, open-source tool that was built to help security teams test their own defenses. It uses several AI agents that work together and can run on almost any AI model. Given a target, these agents can scout the system, find weak spots and steal data with very little human help.
That is what makes it risky. Microsoft says attackers in several regions are now using it, and CrowdStrike linked it to data theft at South Korean banks earlier this month. Storm-0940 has been active since at least 2021 and was mostly known for guessing passwords and exploiting exposed apps. ARTEX lets the group do the same things much faster.
What the Agents Did Once They Got In
In the accounts they managed to enter, the AI agents explored the system, changed app credentials so they could stay inside, and edited user groups and permissions. In one case, a hacked app gave itself the top Global Administrator role. A foothold like that can stay in place even after the original stolen passwords are changed.
Targeted Does Not Mean Breached
The number 2,250 counts accounts the hackers tried to get into, not accounts they actually broke into. Microsoft says most attempts failed. A separate password-guessing campaign between October 10 and 14 saw more than 151,000 failed logins and not a single success.
Some attacks did work, though. The Next Web reported that security firm Huntress confirmed 162 hacked accounts by October 16. It later found more than 300 successful attempts among nearly 1,000 accounts that had been probed. ConnectWise shut down seven app identities on October 17, took Cloud Backup offline, and planned to start bringing partners back on October 26.
Steps Microsoft Wants Companies to Take
The weak spots here are not new. Vendor apps were trusted too much, they had too many permissions, and most monitoring watched people rather than apps. What changed is the speed. AI agents don’t get tired, and they can try many routes at once.
Microsoft’s advice for ConnectWise customers is simple:
- Keep ConnectWise app connections switched off until the company shares recovery steps.
- Save Entra sign-in and audit logs before deleting anything.
- Check account activity from October 15 onward.
- Review every third-party app that can access the account, and what it is allowed to do.
The Danger of Handing Hackers an AI Workforce
The ConnectWise attack may be an early warning rather than a one-off. Until now, a campaign this wide needed a large team of skilled people working for weeks. With tools like ARTEX, a small group can send out hundreds of tireless AI agents that scan, guess, adapt and dig in at the same time, in any language, around the clock. If the same approach is turned on hospitals, power grids, banks or government systems, the damage could spread faster than human defenders can react. The tools are also free and open to anyone, so the skills once held by a few state-backed teams could soon reach criminal gangs and lone attackers. Security teams will likely need to answer with their own AI-driven defenses, because a person reading alerts one at a time cannot keep pace with agents that never stop working.
