An AI-assisted campaign involving ConnectWise has put cloud-application permissions under renewed scrutiny. TechCrunch covered Microsoft’s findings on October 22.
The primary account contains an important distinction: attempted access to a large number of environments is not confirmation that all of them were compromised. For defenders, the story concerns both the speed of automation and the authority already granted to a trusted application.
What the campaign numbers actually measure
In its October 21 research report, Microsoft attributes the activity to Storm-0940, assessed as China-state aligned. It reports ARTEX-assisted sign-in attempts against more than 2,250 tenants following unauthorized access associated with ConnectWise Cloud Backup.
Microsoft separately describes an October 10–14 password-spray campaign peaking at 2,900 failed sign-ins per minute, without observed successful authentication. That is not evidence that 2,250 organizations were breached at that rate.
The report identifies follow-on credential and privilege changes in multiple environments, but says an affected application or successful authentication does not necessarily establish data access or theft. Attribution, attempted entry, established access and exfiltration are different findings, each requiring its own evidence. An incident summary should therefore state what each number counts, the period it covers, and whether its denominator is accounts, tenants or requests. Without those definitions, comparing campaign scale can become misleading even when the individual figures are accurate.
Restoring the vendor service is a separate track
ConnectWise’s official advisory, updated October 23 at 01:00 UTC, says its investigation and restoration work continue. It anticipates completing testing and security validation by the weekend’s end, subject to success, before returning capabilities in phases. It expects access for restoring existing backups early the following week, followed by phased re-enrollment. These are conditional expectations, not a declaration that normal service has resumed.
The company says stored archives remain encrypted, isolated and preserved, with no indication of compromise. It also emphasizes that reviewing individual Microsoft 365 environments remains necessary and tells partners to keep affected connections disabled until applicable formal instructions are issued. Its statement about stored backup archives should not be treated as a clean bill of health for every live customer environment.
This creates two different recovery questions. One concerns whether a backup service can operate safely again. The other concerns what an attacker may have changed after reaching a customer’s environment. Completing the first does not logically complete the second. A useful recovery report would say which environment was examined and what evidence supports its conclusion, rather than letting a vendor-wide status label stand in for a customer’s own investigation.
Automation amplifies an existing trust problem
The distinctive AI element is accelerated execution, not evidence that an agent created new permissions from nothing. The reported activity used application relationships that already carried authority. That makes the scope of delegated access as important to the analysis as the speed of the software using it.
An investigation would consequently need to distinguish an application’s intended work from changes made through its credentials. A list of connections alone cannot answer that question. Nor can a service-availability update show whether an unauthorized identity or permission persists elsewhere.
There is also a measurement problem in describing campaigns at machine speed. A very high count of failures may demonstrate aggressive automation while saying little about successful access. A smaller set of successful actions may have greater operational significance. Reporting both under the same label can obscure where the real harm has been established.
The current evidence supports a serious supply-chain and delegated-access incident, with an AI-assisted component attributed by Microsoft and ongoing remediation described by ConnectWise. It does not support treating every targeted tenant as a confirmed data-theft victim. That boundary should remain visible as the investigation and restoration proceed.