Recorded from the investigating organization's published analysis. Figures below are the figures stated in that analysis. The attribution to a named actor is Microsoft's assessment and is recorded as such. ConnectWise published its own initial advisory on 16 October 2026.
The application was trusted. The vendor was compromised.
Microsoft attributes the compromise of a ConnectWise multi-tenant application to Storm-0940, an actor it assesses as China-state aligned and has tracked since 2021. From September 2026 the actor used ARTEX — an open-source multi-agent framework that runs reconnaissance through to exfiltration with little human interaction — to automate attacks against hundreds of organizations.
Having obtained control of the identities of several ConnectWise cloud applications, the attacker authenticated into customer tenants on the trust those applications already held. No interactive user sign-in was required, because none was ever required. In some environments the activity went further: directory operations, application credential changes for persistence, changes to identities, groups and privileged roles, and bulk reads of secrets across multiple key vaults.
Attempted sign-ins to more than 2,250 tenants. Microsoft records over 200 sign-ins in under four minutes using four application credentials.
Several attacks in the wider campaign reached privileged cloud access within three hours of initial targeting.
A separate password-spray run from 10 to 14 October produced over 151,000 failed logons against 327 accounts, peaking near 2,900 failures a minute. Microsoft records no successful authentication from it.
Indicators matching the same infrastructure were surfaced in public reporting on 15 October. The vendor's advisory followed on 16 October, and later confirmed that one address matched the spray infrastructure.
Unlike the evaluation incidents recorded elsewhere in this register, this entry describes a deliberate intrusion by an identified actor. The attribution is Microsoft's published assessment. Zevana does not extend it.
The trust was standing, and nobody was present when it was used. A multi-tenant application holds consented access that does not require a user to sign in. That is the feature it is bought for. It also means that compromising the vendor is functionally equivalent to compromising every tenant that consented — and the tenant finds out from somebody else.
Two controls failed behind it. Scope was granted once and never reduced. Permissions sized for integration day remained in force long after, wide enough to reach identities, privileged roles and vault secrets from an application bought to do something narrower. And response was built for human pace. Over 200 sign-ins in under four minutes, privileged access inside three hours: an alert queue reviewed each morning does not intersect with that timeline at any point.
The transferable finding is about the direction detection travelled. Indicators appeared in public reporting before the vendor's own advisory, which means the tenants carrying the exposure learned of it last. A control that depends on your supplier telling you in time is not a control you hold.
Exposure floors are recorded, not averaged down.
Any organization that has consented a third-party application into its identity environment carries this structure. The analogue is heightened because the exposure is inherited rather than chosen: it was created by a supplier's compromise, it operates through access the organization itself approved, and it is invisible from inside the tenant until somebody outside says so.
Governance & policy · Risk & control readiness
Atlas establishes who may grant an application standing access, on what basis, and who reviews it afterwards. Strata addresses the control side: permission scope matched to purpose, a current inventory of consented applications, vendor review, and detection that does not wait on a supplier's disclosure.
Which third-party applications can act inside our systems without a person signing in?
Ask for the list, not the assurance. If the list takes more than a day to produce, that is the finding — the access exists either way, and the inventory is what makes it governable.