Microsoft named the tool behind the campaign. Its 21 October research post attributes the ConnectWise Cloud Backup supply-chain compromise to the China-state-aligned actor Storm-0940, active since 2021 with no known industry aliases, and says the actor “used ARTEX to attempt to sign-in to more than 2,250 tenants”. ARTEX is an open-source agentic AI penetration-testing framework. The operative verb in that sentence is attempt to sign-in: Microsoft is not reporting 2,250 compromised organisations, and it adds that “the majority of the targeting wasn’t successful”.

The compromise was public before that post, and the order of events matters to how the figures are read. Microsoft’s account says public reports of the ConnectWise compromise started to surface on 15 October, the day the ConnectWise application was used to sign in to customer Microsoft 365 tenants. ConnectWise published its first advisory on 16 October at 17:47 UTC, and Huntress published tenant-level confirmation the same day. The 21 October post added the framework’s name, Microsoft’s telemetry and the figures that have circulated since.

Microsoft assesses that Storm-0940 used ARTEX in the ConnectWise compromise

The attribution is Microsoft’s, and the post frames it as an assessment: “Microsoft assesses that Storm-0940 was responsible for the recently disclosed ConnectWise supply chain compromise of a multi-tenant application.”

ConnectWise’s advisory states a narrower conclusion in its own words. Its 21 October update reads: “Based on the evidence available to ConnectWise, including threat intelligence shared by Microsoft, it appears that a China state-aligned group was responsible for the attack on ConnectWise Cloud Backup. The group used an openly available agentic AI framework to automate and speed up its operation.” That attribution rests on Microsoft’s intelligence rather than investigative work of ConnectWise’s own, and the advisory does not name Storm-0940. The same advisory says ConnectWise “contained the affected application connections” and that on 19 October it “completed a full reset of all accounts and credentials in the affected Cloud Backup environment”.

No independent lab has separately named Storm-0940 for this incident. Wiz, which is tracking an ARTEX-using actor, describes its observations as “overlapping with Storm-0940” and labels the relevant indicators as “Microsoft attributes to Storm-0940”. Wiz counts two separate actors tied to the framework, noting that “Microsoft and Zscaler have since tied it to two separate actors”: Storm-0940, and a second, unattributed operator behind the Korean financial-sector campaign. CrowdStrike, Recorded Future and Zscaler describe that same operator, assessed as Chinese-speaking and financially motivated, and Recorded Future tracks it as TAG-217.

Why the spray figures and the 2,250 attempted sign-ins are not interchangeable

Microsoft’s post describes activity in two phases, and the figures from each do not transfer to the other.

The first is a cross-tenant password spray from 10 to 14 October. Microsoft’s telemetry covers 327 accounts across eight organisations, more than 151,000 failed logon events, a peak of about 2,900 failures per minute on 11 October, and lockout responses in nearly 98% of activity. In that phase Microsoft observed “no successful authentication, token issuance, Graph call, or directory change”. In Microsoft’s telemetry the spray produced no tenant access at all, and 2,900 per minute is a count of failed logons at its peak.

The second phase runs from 15 October, when the actor authenticated to Microsoft 365 customer environments through ConnectWise applications that had already been consented to. Those applications sign in as themselves with permissions the customer had already granted, so no interactive user sign-in and no MFA prompt is involved, and whoever holds the credential inherits that access in every tenant where the app is installed. This is where the “more than 2,250 tenants” attempted sign-ins belong. Where the activity succeeded, Microsoft observed Microsoft Graph reconnaissance, modification of application credentials to establish persistence, a self-assignment to Global Administrator from one of the ConnectWise applications, and bulk secret reads from Azure Key Vaults with key enumeration and a key-unwrapping operation.

Huntress, which had tenant logs, confirmed unauthorised sign-ins in 162 tenants as of 16 October, then identified almost 1,000 Entra tenants with sign-in activity from one IP address by 17 October, with more than 300 succeeding. Those totals were still being revised.

A third Microsoft figure covers the campaign as a whole rather than either phase: “multiple instances where the time from initial targeting to privileged access was less than 3 hours”. Microsoft describes this as multiple instances across a campaign that began in September, so it is neither an average nor a figure for every tenant, and it is not limited to the ConnectWise phase.

The Register’s 22 October headline says the breach “let Chinese spies loose on 2,250 cloud tenants”. That is the outlet’s characterisation of the event, and it reads as access or victimhood. The Register’s own body keeps Microsoft’s verb, reporting that Storm-0940 “attempted to authenticate against more than 2,250 Microsoft 365 tenants”. The same piece moves the 2,900-per-minute figure from the 11 October spray onto the later app phase, where Microsoft does not place it.

No published baseline supports “exponential” or “machine speed”

Microsoft states that the activity “demonstrates how threat actors are integrating agentic AI into existing offensive security workflows and allows them to scale their operations exponentially”, and says the actor used ARTEX “to accelerate the activity to machine speed”. ConnectWise repeats the machine-speed phrasing in its advisory. Both are comparative assertions about what an agent can do, and no source in this story publishes a human-operated comparator to measure them against.

Recorded Future’s Insikt Group examined a separate ARTEX-using actor and reached the most direct published conclusion: “The available evidence does not establish the extent to which AI accelerated these attacks”, adding that “differentiating between agentic and human-operated attacks can be difficult”. Wiz corroborates a fast, coordinated cadence in its own telemetry, with a median gap of one second between authentication attempts and some sessions lasting 24 to 30 hours, and still concedes that “none of the individual techniques observed are novel”. Huntress measured a downstream tenant sequence that “took about five minutes”, but it timed a post-compromise procedure, not Microsoft’s initial-targeting-to-privileged-access interval.

There is also no clean non-AI comparison in this campaign. Microsoft does attribute the spraying to ARTEX: “During the ConnectWise supply chain compromise, Storm-0940 used ARTEX to accelerate scanning and reconnaissance, password spraying, and OAuth application abuse at downstream victims.” What the post does not publish is a measure of how much of the campaign’s speed or success the AI contributed. Microsoft’s own history for the actor complicates the simpler reading: Storm-0940’s playbook since 2021 already relied on password spraying and exploitation of internet-facing applications, so the baseline technique predates ARTEX. Naming the framework does not, by itself, show how much of the outcome the AI produced.

What ARTEX is, and which other actors have used it

Microsoft describes ARTEX as “an open-source penetration testing framework that can use any AI model to automate a variety of offensive security workflows”, and more specifically as “a multi-agent LLM system written in GO with a Next.js frontend”. Its interface and output default to English and can switch to Korean, Mandarin, Spanish or Japanese, and its agents “run an autonomous penetration test from reconnaissance through data exfiltration with little needed human interaction”.

Wiz traced the tool’s path. A Chinese developer published it on GitHub in July 2026, it won Baidu’s “Agent+” red-team challenge that September, and the developer made it closed source on 8 October, by which point it had more than 12,000 forks. Wiz recorded ARTEX artifacts in attacks on South Korean banks in late September and on Japanese companies in October. Microsoft notes the same public reporting and says threat actors across multiple regions use the framework, without attributing the Korean breaches to Storm-0940.

No published confirmation that customer data was accessed

The gap that changes how a reader should read the headline is the absence of any confirmation that customer data was accessed or exfiltrated in the ConnectWise incident. Microsoft cautions that “the presence of an affected application or successful threat actor authentication does not necessarily indicate that customer data was accessed or exfiltrated”. Huntress found no mail or file access in the one tenant it audited in detail, and noted that directory reads through Microsoft Graph leave no audit trail. Microsoft’s post also tells defenders to inventory “all four application IDs” but names three in its body and hunting queries, so the fourth is not public. ConnectWise has published no victim count and no forensic findings of its own, and says it is “working with independent third-party specialists to determine how the unauthorized access first occurred and its full scope”.