AI agents tried 2,250 tenants through one backup vendor
Microsoft said a China state-aligned group used an open-source AI agent framework after breaking into ConnectWise's cloud backup service, then tried to sign in to more than 2,250 customer tenants. ConnectWise has suspended the service and says its backup archives show no sign of compromise.
A China state-aligned group used AI agents to try to sign in to more than 2,250 Microsoft 365 customer tenants after compromising ConnectWise's Cloud Backup service, Microsoft said on Wednesday. It put the pace at 2,900 sign-in attempts a minute.
Microsoft tracks the group as Storm-0940 and says it has been active since 2021, stealing data from organisations in many industries. It assessed that the group was behind the ConnectWise compromise. ConnectWise, which sells IT management software mainly to managed service providers, has suspended Cloud Backup, a service formerly called SkyKick that copies customers' Microsoft 365 data. A tenant is one organisation's own Microsoft 365 environment.
The route in was the trust customers had already given the vendor. After taking control of the identities of several ConnectWise cloud applications, Microsoft said, the attackers' agents used those app credentials to authenticate to customer environments, with no interactive user sign-in needed. In some tenants they went further, into reconnaissance through Microsoft Graph, changes to application credentials that can act as a backdoor, and changes to identities, groups and privileged roles. Microsoft said it found "multiple customer environments" with such follow-on activity, but cautioned that an affected application, or even a successful sign-in, "does not necessarily indicate" that data was accessed or taken. TechCrunch reported that the credentials could let the hackers reach emails and other files.
The tool was ARTEX, which Microsoft describes as an open-source penetration-testing framework of AI agents that can use any AI model and run an attack from reconnaissance to data theft with little human input. Storm-0940 has used it since at least September against organisations in sectors from aerospace and defence to think tanks, Microsoft said, in several cases reaching privileged cloud access less than three hours after first targeting. Microsoft also noted public reporting this month that ARTEX was used in data breaches at Korean financial institutions. Agents "are not constrained by factors affecting humans", it wrote: they work around the clock and try several paths at once.
Not everything worked. Between 10 and 14 October, Microsoft said, infrastructure linked to the group ran an automated password-spray campaign against 327 accounts at eight organisations, logging more than 151,000 failed sign-ins, over 111,000 of them on a single day. Lockouts made up nearly 98 per cent of the activity, and Microsoft saw no successful authentication. ConnectWise's own advisory says it discovered the intrusion on 14 October and first published a notice on Friday 16 October. On 17 October it shut off Cloud Backup access, disabled seven ConnectWise application identities and revoked the certificates used by the affected connections; by 18 October it had also suspended two linked products, SaaS Security and Cloud Management. It reset all accounts and credentials in the affected environment on 19 October and began a phased restoration on Wednesday, which it expects to take several days.
In its update on Thursday, ConnectWise said: "Existing backup archives remain encrypted, isolated and preserved, with no indication of compromise." It added that it has "no indication that other ConnectWise products, services or offerings are impacted by this incident". It has said its systems cannot show whether individual customers' Microsoft 365 tenants were affected, so each must be reviewed. Those are the company's statements, and Komposite has not independently verified them.
Microsoft's advice is specific: keep the affected ConnectWise connections disabled until ConnectWise issues formal restoration instructions, and do not delete the enterprise applications as a first response, so that evidence survives. It told customers to search every tenant for the applications by ID rather than display name, and to review service-principal sign-ins, the log of app-only logins, from at least 15 October, or from 1 October where records allow.
Why it matters: the entry route, a vendor's stored credentials, is familiar; the speed is not. Microsoft's figures describe an automated sweep across thousands of tenants that left defenders little time to respond, even though most of the targeting failed. As we reported, shared vendors concentrate this kind of risk, and Railway's warning on Wednesday that every customer should rotate its keys is another case of one platform holding many companies' secrets.