Case 004 · Incident: 13–15 Oct 2026
ConnectWise: China-linked hackers let AI agents run the break-in
In mid-October 2026, an attacker used a compromised credential belonging to ConnectWise Cloud Backup (formerly SkyKick), a multi-tenant Microsoft 365 backup app used by MSPs, to get Microsoft Graph tokens in customer tenants. Huntress confirmed unauthorized sign-ins in 162 tenants, a figure later reported as 173. In a few tenants, ConnectWise Cloud Management apps gave themselves Global Administrator and planted secrets on other apps so the attacker could keep access.
A Secret Set to Expire in 2028
In one compromised tenant, two applications, one of them the security firm Huntress's own platform, were given new client secrets. A client secret is a kind of password for an app. Help Net Security reported that those secrets were set to stay valid until 2028, a way back in that would survive the shutdown of ConnectWise's apps, though who planted them is still not settled.[1][2][3] By October 17, 2026, ConnectWise had revoked the certificates behind its Cloud Backup service and switched it off. Scheduled backups stopped. Archive access and restores went dark for the providers that depended on it.[4] This hit in the middle of October 2026. The security firm Huntress raised the alarm publicly on October 16. It confirmed unauthorized sign-ins in 162 tenants, a figure later reported as 173.[1][3][5]
The victim was ConnectWise Cloud Backup, formerly SkyKick, acquired in 2024. It is a cloud backup and recovery service built for managed service providers to protect their clients' Microsoft 365 or Google Workspace data.[1][2]
One Key, Many Doors
To do its job, the app is granted broad access in every tenant where it is installed. That means mailboxes, OneDrive, SharePoint, Teams and directory data.[1] ConnectWise says more than 100,000 managed service providers use its platform. Each one may look after dozens or hundreds of client organizations.[2] That is the appeal for an attacker. As ConnectWise itself put it, one trusted connection can open the door to many organizations.[6]
Rewind: The Way In
Rewind to the days before. Between October 10 and 14, 2026, infrastructure tied to the attacker ran an automated password-spray campaign, targeting 327 accounts across eight organizations.[7] Microsoft's telemetry captured more than 151,000 failed logons. The activity peaked on October 11 at 2,900 failures per minute, with over 111,000 failures crammed into a single day.[7] 48 accounts took more than 1,000 failures each. Account lockouts made up nearly 98% of the activity. Despite the scale, the spray produced no successful sign-in.[7]
Huntress saw a second address spray four law firm tenants on October 13 and 14, with no successful user sign-ins detected. That address matters: it is the one that would later run the token sweep. Microsoft's spray came from a different address in the same range. Loud, and unsuccessful.[1][7] As far as investigators can tell, the sprays did not get the attacker in, because no successful sign-ins were seen. What was used was a credential belonging to Cloud Backup itself, called a client assertion, a signed proof an app uses to show who it is. It lives in the vendor's own tenant, so nothing a customer did could have produced it.[1][3][7] How that credential was exposed is still unknown. ConnectWise says it is unclear whether a product vulnerability played any role, and the exact exposure mechanism is still under analysis.[4]
On October 15, between 06:13 and 08:34 UTC, the stolen credential was used from that same single address to get Microsoft Graph tokens inside customer tenants. Graph is the main interface for reading and changing data in Microsoft's cloud. A token is a short-lived pass to use it.[1] The actor moved through tenants in the sorted order of their unique identifiers, a few seconds apart, which points to a script working from a target list. BleepingComputer reported that because the requests came from an automated script and involved no user login, multi-factor authentication, the extra check such as a code on your phone, was ineffective.[1][8] Huntress watched the Cloud Backup identity sign in to 162 tenants between October 15 and 16, a confirmed count later reported as 173. Across almost 1,000 tenants in Entra, Microsoft's identity service, Huntress found sign-in activity from the attacker's address, and 300 of those sign-ins succeeded.[1][3][4]
In some tenants, the sign-in succeeded but no ConnectWise service principal existed. A service principal is the local identity an app gets inside a tenant. Its absence is said to have prevented the attacker from gaining access to those tenants.[4] Huntress also reviewed application sign-ins across more than 10,000 tenants with ConnectWise apps, where it could see those logs. Only that one attacker address and Microsoft-internal ranges produced application sign-in activity.[9]
Rewind: Digging In
In at least one tenant, something else happened, and the timing is odd. The ConnectWise Cloud Management app for Microsoft Teams added itself to the Global Administrator role, the highest level of control in a Microsoft cloud tenant, and SecurityWeek reported that this came hours before the token sweep began. Who was behind that step is not confirmed.[1][3] Then a Python script modified two other applications in the tenant: an MSP management tool, and Huntress's own Security Platform. Both received new client secrets.[1] Help Net Security reported that those planted credentials were set to remain valid until 2028. According to that report, they gave whoever created them a way back in that would survive the shutdown of ConnectWise's apps.[2]
Huntress saw the same pattern in a second tenant: self-granted Global Administrator, then planted secrets. Which apps got backdoored appeared random in the two tenants it reviewed.[9] In nearly 50 monitored tenants, Cloud Management service principals assigned themselves directory roles, mostly Exchange Administrator. One Reddit user, whose report has not been verified, said Microsoft's Privileged Identity Management, a service that watches powerful roles, alerted them on October 15 that a SkyKick service principal had assigned itself that role. Whether these self-assignments were ConnectWise's own automation or the attacker is still unresolved.[2][8] Two more details make this harder to read. SecurityWeek reported that the Global Administrator activity came from Microsoft's internal cloud network, not from the attacker's address. And Help Net Security reported that the software signature on the Exchange Administrator self-assignments matches the one ConnectWise's own apps use. That is a hint, not a confirmation.[2][3]
Microsoft described the follow-on activity as reconnaissance through the Graph interface. It also saw attempts to add credentials to the related service principal in target tenants, to keep direct access as a backdoor.[1]
Who, and How Fast
On October 21, 2026, Microsoft named the actor. It assesses that Storm-0940, a China-state aligned group active since 2021, was behind the compromise. The group targets many industries for data theft.[7] According to The Record, Microsoft said it tied the campaign to Storm-0940 because IP addresses published by ConnectWise matched ones seen in its own investigations.[10] The Hacker News reported that Microsoft first described Storm-0940 in November 2024. At the time, the group was leveraging a botnet, a network of hijacked devices, called Quad7 to conduct highly evasive password spray attacks.[6][7]
What was new was the tooling. Storm-0940 used ARTEX, an open-source penetration testing framework that can plug in any AI model to automate offensive work. Here it accelerated scanning, reconnaissance, password spraying and OAuth app abuse.[7] Microsoft says that following the unauthorized access, Storm-0940 used ARTEX to attempt sign-ins to more than 2,250 tenants. Those were attempts, a separate and larger figure than the confirmed counts.[7] Microsoft says that in some environments, the AI agents went beyond logging in. They moved on to reconnaissance, app credential changes for persistence, and changes involving identities, groups and privileged roles. Microsoft warns those changes can stay active as a backdoor even after the original app is cleaned up. Huntress separately found self-granted Global Administrator and planted secrets in two tenants. But it has not tied that activity to the same operator as the sign-in sweep.[3][7][9]
Microsoft also recorded the pace. In Storm-0940's wider ARTEX campaign, there were multiple instances where the time from initial targeting to privileged access was less than 3 hours.[7] The Record identified the developer of ARTEX as Chinese cybersecurity engineer Li Puhua. According to The Record, the developer said in a statement on GitHub on October 8 that the tool had been abused by malicious actors. The developer added that the project would no longer be updated and would be converted to closed source.[10] The Record, citing Reuters, reported that Chinese Foreign Ministry spokesperson Mao Ning denied any knowledge of the incidents. Microsoft, for its part, says the activity did not come from a compromise of, or a vulnerability in, its own systems or products.[7][10]
Was Anything Taken?
Back to the damage. Todyl, another security firm, said the sign-ins gave the attacker the technical ability to read and write mail and attachments, OneDrive and SharePoint files, and Teams chats.[11] Being able to do something is not proof it was done. Across nine tenants, Todyl found only sign-in activity, with no evidence of new accounts, added permissions, changed passwords or persistence through the backup app, and Huntress saw no mailbox, file, user or policy changes from the main sweep, though the access may have been used to read the directory through Graph. That is not the whole picture, because Microsoft reports changes involving identities, groups and privileged roles in some environments, and in two tenants a Cloud Management app granted itself Global Administrator.[1][7][8][9] But absence of evidence is weak comfort here. Huntress cautioned that Microsoft does not record some kinds of data reads by applications, and that some access, such as directory reads through Graph, leaves no audit trail. That is why it urged acting now instead of waiting for proof.[1][2]
ConnectWise says the existing backup archives stayed encrypted, isolated and preserved, with no evidence they were accessed. But it also admits it cannot directly observe downstream customer environments, and Microsoft adds that a successful sign-in does not necessarily mean data was taken.[4][7] One wave was precise. On October 15 a second, targeted effort singled out just 16 tenants in the policy, national security, trade and critical mineral sectors. It failed.[4]
The Response
When did ConnectWise first notice? Even that is contested. ConnectWise says its own monitoring identified the issue on October 14, 2026. Help Net Security reported that ConnectWise discovered the problem on October 15 and has since engaged an outside forensics firm.[2][4] ConnectWise began rolling out mitigations on October 17 at 5:00 a.m. UTC. That was three days after the date ConnectWise gives for first identifying the incident. At 07:00 UTC it disabled seven application identities, and suspended Cloud Backup by revoking the certificates used for those connections.[4][9] Disabling the connections stops new logins. It does not undo changes already made inside a tenant. The shared credentials are dead, but tokens issued earlier may keep working until they expire.[4]
As a precaution, ConnectWise also took SaaS Security and Cloud Management offline while it completed its review. With Cloud Backup suspended, scheduled backups stopped running, and access to backup archives and restores was unavailable.[4] Over the following days the picture widened. ConnectWise shared more linked addresses, brought in independent incident-response specialists, and on October 20 said it had finished an account and credential reset, with a phased restoration expected to take several days.[4][6][9] iTnews reported that Huntress, whose own application had a rogue credential planted on it, said it had found and removed that credential.[5]
Huntress says it has not identified new malicious activity since October 15, though there is no evidence the stop was caused by public disclosure. And this may not be over. The Hacker News reported that Wiz, which is owned by Google, said it observed Storm-0940 expanding its attempted supply chain compromises into other software sectors.[4][6]
What Defenders Should Change
The lesson is structural. As ConnectWise put it, one trusted connection can open the door to many organizations. In a supply chain, your security depends on the vendor holding the key.[6]
What to do
- If a vendor's app in your tenant is compromised, disable it until the vendor confirms every credential has been reset, and turn on Microsoft Graph activity logs so that reads by apps leave a record. Huntress advises auditing for any app credentials added since October 13, 2026. CoreView separately recommends a wider review: mapping and rolling back any change you cannot confirm between October 1, 2026, and 07:00 on October 17, 2026.
- Use Entra application management policies to restrict or block client secrets on app registrations, and to cap how long any credential can live. A credential with a capped lifetime cannot quietly stay valid for years.
- Apply least privilege to every multi-tenant app and every consent grant. An app only needs the permissions its job requires, and broad directory write access should be rare and deliberate.
- If you do find a suspicious app, resist the urge to delete it first. Preserve its configuration and the evidence before you make any irreversible change, so investigators can reconstruct what happened.
- And understand how access actually ends. Revoking user sessions does not stop this kind of app access. Disabling an app halts new tokens, but tokens already issued can stay valid for a short window after. Plan your response around that gap.
Open questions
- How was the Cloud Backup credential exposed, and did a product vulnerability play a role? ConnectWise says this is still under investigation.
- Was any customer mailbox, file or Teams data actually accessed or exfiltrated? Some Graph reads by applications leave no audit trail.
- How many tenants were really affected? Sources give 162 confirmed, 173 confirmed, about 300 successful sign-ins out of almost 1,000 tenants, and more than 2,250 tenants targeted.
- When did ConnectWise first detect the incident? Sources say October 14 (ConnectWise via THN), October 15 (Help Net Security) and October 16 (ITPro).
- Was the Global Administrator/secret-planting activity (from Microsoft's internal network, about 2.5 hours before the main sweep) done by the same operator? Huntress has not linked the two phases.
- Were the Exchange Administrator self-assignments in about 50 tenants ConnectWise's own automation or the attacker? ConnectWise has not confirmed.
- Sources disagree on whether the Cloud Management Exchange Online, Teams and SharePoint apps were disabled early (D2 says they appeared not to be; D13 says they were shut down).
- What was the attackers' goal in the ConnectWise operation? CoreView says no clear reports of the purpose exist.
- When will Cloud Backup, SaaS Security and Cloud Management be fully restored?
- How many planted backdoor credentials remain in downstream MSP and customer tenants?
Sources
- [1]The Hacker News — Hackers Use Compromised ConnectWise Credential to Sign In to 173 Microsoft 365 Tenants
- [2]Help Net Security
- [3]SecurityWeek
- [4]The Hacker News — ConnectWise Suspends Cloud Backup After Attackers Compromise Microsoft 365 Tenants
- [5]iTnews
- [6]The Hacker News — China-Linked Hackers Use ARTEX AI Agents in ConnectWise Supply Chain Attack
- [7]Microsoft Threat Intelligence — AI-driven, large-scale nation-state campaign using ARTEX enables supply chain compromise
- [8]BleepingComputer
- [9]The Hacker News — ConnectWise Resets Credentials as Cloud Backup Remains Offline After Security Incident
- [10]The Record — Microsoft blames ConnectWise attacks on Chinese state hacker using ARTEX AI agents
- [11]ITPro
- [12]CoreView