China-linked hackers used AI agents to target 2,250+ Microsoft tenants
What's the deal? Microsoft disclosed that Storm-0940, a China-state aligned actor, used an open-source AI penetration-testing framework called ARTEX to automate attacks on hundreds of organisations worldwide. Starting as early as September 2026, the agents ran intrusions from reconnaissance through data exfiltration with little human input.
How it worked: Microsoft links Storm-0940 to the ConnectWiseDealroom has a profile for this one. Try Dealroom → supply chain compromise. After gaining unauthorised access to ConnectWise's Cloud Backup application, the actor used ARTEX to try to sign in to more than 2,250 Microsoft 365 customer tenants at 2,900 attempts per minute.
The 2,250 figure refers to tenants targeted, not confirmed breaches. In a separate password-spray campaign between October 10 and 14, Storm-0940 hit 327 accounts across eight organisations, generating more than 151,000 failed logon events — with no successful authentication observed in that phase.
Why it matters: In some cases, the time from initial targeting to privileged cloud access was under three hours. Where agents got in, they moved to reconnaissance, modified application credentials for persistence, and changed identities, groups, and privileged roles. Targets spanned aerospace and defence, automotive, energy, financial services, government, semiconductors, software and think tanks, among others.
ARTEX is a multi-agent framework written in Go with a Next.js frontend that can run on any AI model. In October 2026 it was also reported in use against Korean financial institutions in separate data breaches.
What could go wrong? The attacks exploited trust ConnectWise's apps already held inside each tenant, not a breach of Microsoft itself. ConnectWise published an advisory on October 16 and is urging customers to keep affected application connections disabled until formal restoration instructions arrive.
The bigger picture: The campaign fits a wider collapse in the time defenders have to react. Data from ZeroDayClock, highlighted by Andreessen Horowitz, shows that about 87% of exploited vulnerabilities are now attacked on or before the day they become public, up from 23% in 2020, with a median time to exploit of just one day. As AI speeds up both finding and exploiting flaws, investors have rewarded cybersecurity stocks with higher revenue multiples than the rest of software.
The signal: Microsoft says this shows threat actors folding agentic AI into offensive workflows to scale operations, since agents work around the clock and attempt multiple breach paths at once. The weak points remain familiar — compromised app credentials and excessive cloud privileges — but AI is making the repetitive work of finding and exploiting them far cheaper and faster.
Read more: Microsoft Security Blog · a16z Charts of the Week
