TL;DR: On November 13, 2026, the FBI issued FLASH-20261113-001, warning that attackers are using ARTEX, an open-source AI hacking framework, to run intrusions at machine speed. One campaign went through ConnectWise Cloud Backup and opened a path into more than 2,250 customer Microsoft 365 tenants. If your organization or your MSP ever connected a ConnectWise app to Microsoft 365, check your tenant now. The logs you need are close to aging out.
This isn’t a story about a brand-new kind of bug. The attackers found ordinary weaknesses (exposed credentials, weak passwords, trusted app connections) and let an AI agent chain them together faster than people could respond. The FBI says some intrusions went from first targeting to privileged access in under three hours.
What happened
Attackers compromised the identities behind ConnectWise’s Microsoft 365 integrations, then used that trusted access to reach the tenants of ConnectWise customers. Microsoft says the campaign began as early as September 2026, and the FBI ties ARTEX activity in general back to at least August.
activity in general back to at least August.
Date (2026) | Event |
|---|---|
July | A developer called “Autumn-27” publishes ARTEX on GitHub as a pentesting tool. |
August | The FBI says threat actors start using ARTEX in real campaigns. |
September | The ConnectWise campaign begins, per Microsoft. |
Oct 7 | CrowdStrike reports ARTEX being used against South Korean financial firms. |
Oct 8 | The developer pulls the GitHub repository. Forks and mirrors keep it alive. |
Oct 10–14 | Password spraying hits ConnectWise-linked identities, peaking Oct 11. |
Oct 15 | ConnectWise compromise is first reported publicly. |
Oct 16 | ConnectWise publishes its advisory. |
Oct 21 | Microsoft publishes its analysis and customer guidance. |
Early Nov | Researchers count more than 1,600 ARTEX consoles exposed on the internet. |
Nov 13 | The FBI issues FLASH-20261113-001. |
The FBI’s wording matters here. It says the breach “enabled access to” more than 2,250 downstream tenants. That confirms the attackers could get in. It doesn’t confirm that data was stolen from every one of them, and Microsoft makes the same point.
Who was involved
Party | Role |
|---|---|
Storm-0940 | The name Microsoft uses for the group behind the ConnectWise campaign. Microsoft calls it a nation-state campaign. The FBI advisory names no country or group. |
ConnectWise | The supply-chain link. Its Cloud Backup, Teams and Graph integrations were the trusted apps the attackers abused. |
ConnectWise customers and MSPs | More than 2,250 Microsoft 365 tenants that were reachable through those apps. |
Microsoft | Investigated the campaign, published indicators and hunting guidance, and added Defender detections. |
FBI | Pulled the research together into a TLP:CLEAR FLASH advisory. |
South Korean financial institutions | A separate ARTEX campaign that exposed personal data on about 67,000 people. |
Security researchers | CrowdStrike, Zscaler, Google Threat Intelligence and Mandiant, Akamai, Censys, Check Point and CERT-EU all published research the FBI cites. |
News coverage has been loose with some of these details. Some coverage says the FBI attributed the attack to China, or lists Barracuda as a victim. Neither claim appears in the advisory.
The tools
ARTEX is an open-source, multi-agent penetration testing framework written in Go. You give it a goal in plain English, and its agents turn that into commands, from reconnaissance through data theft. It has no AI model of its own. It plugs into outside LLMs: Anthropic Claude, OpenAI GPT, and open-source or Chinese models such as DeepSeek, GLM, Grok and Kimi. In attacker-run instances the FBI looked at, cheap Chinese models, mostly DeepSeek Flash, were the most common.
The developer took the repository down on October 8, but ARTEX lives on through forks, mirrors, archived copies and a paid “ARTEX PRO” version.
Around ARTEX, the FBI lists this supporting toolkit:
- Remote access tools: WISP, vshell and AdaptixC2, plus a component called BLOODALCHEMY
- Web shells: JSP and PHP shells, some planted through MySQL log poisoning
- Tunneling and hiding: Chisel, SoftEther, Tor, SOCKS proxies and residential proxy services
- Exfiltration: command output sent to rotating webhook.site endpoints, and 100GB cloud volumes created to stage stolen data
- LLM resellers: reseller and gateway services such as xcai[.]pro, CLIProxyAPI and New API to get model access
The tactics are mostly familiar. What’s new is the speed. The FBI notes ARTEX is noisy and does set off alerts, but the attacks move so fast that defenders have very little time to act on them.
How to tell if you’re impacted
You can say for sure whether you were exposed. A clean search can’t prove you weren’t compromised. Microsoft and the FBI both say empty results don’t rule out access, so treat what follows as a way to find evidence, not to clear yourself.
Step 1: Check whether you were exposed
In every Microsoft 365 tenant you own or manage, look for these ConnectWise application IDs. Include tenants where the integration is no longer used. Search by application ID, not display name (Entra admin center → Enterprise applications → filter by Application ID).
Application ID | App |
|---|---|
506e08f9-6e9a-4da4-942b-a099a03644ae | ConnectWise Cloud Backup |
24205314-2b89-4e07-a2a4-579e898a5640 | ConnectWise – MS Teams |
0de14c77-b9b0-4137-9360-34878a929827 | ConnectWise – MS Graph |
Microsoft’s guidance says to check four app IDs but lists only three. Get the fourth from ConnectWise’s Trust Center advisory before you treat this check as complete. If none of these apps was ever in a tenant, this attack path didn’t reach it.
Step 2: Check your log coverage today
The FBI says to review from at least October 1, 2026. That’s more than 45 days ago. By default, Entra sign-in logs are typically kept about 30 days with P1/P2 licenses and 7 days on the free tier, and Microsoft Graph activity logs exist only if someone turned them on beforehand. Unless your logs go to a SIEM or Log Analytics, the key evidence may already be gone. In that case you can’t verify either way, and the safe choice is to treat the tenant as compromised.
Step 3: Hunt for the indicators
Look for activity from these sources. Treat them as a starting point and also look for anything unfamiliar, because attacker infrastructure changes.
Indicator | Context |
|---|---|
144.172.64.60, 144.172.64.99, 144.172.64.128 | Password spraying, sign-ins and Graph API activity |
187.15.88.186, 187.15.89.186, 187.15.90.217 | Transient VPN addresses used Oct 15–16 |
Python-urllib/3.12, python-requests/2.31.0, python-requests/2.34.2, curl/7.88.1, azure-toolkit/1.0 | User agents (from Microsoft) |
This Sentinel query lists sign-ins by the ConnectWise apps. Review successful and failed sign-ins separately.
AADServicePrincipalSignInLogs
| where TimeGenerated >= datetime(2026-10-01)
| where AppId in (“506e08f9-6e9a-4da4-942b-a099a03644ae”,
“24205314-2b89-4e07-a2a4-579e898a5640”,
“0de14c77-b9b0-4137-9360-34878a929827”)
| summarize count() by AppId, IPAddress, ResultType, ResourceDisplayName
Then check these log sources for changes made after a break-in:
- Entra audit logs: new credentials on apps or service principals, new apps or owners, app-role assignments, consent grants, privileged role changes, and changes to users, groups, authentication methods or Conditional Access
- Microsoft Graph activity logs: requests from the ConnectWise app IDs or the IPs above
- Purview audit logs: reconnaissance, mail and file access, downloads, forwarding, inbox or transport rules, and new mailbox delegation
- Azure Key Vault: bulk secret reads across several vaults
- Azure Storage and ARM: any activity from the IPs above
Step 4: Check Defender alerts
Microsoft added detections for this campaign. Alerts such as Unusual Microsoft Graph activity associated with Compromised ConnectWise Application or Application self-assigned Global Administrator role are close to a confirmed compromise.
If your environment goes beyond Microsoft 365
The FBI’s broader ARTEX indicators are worth a sweep too:
- Unexpected .aspx or .jsp files in web application folders
- Unusual Kudu/SCM API activity in Azure App Service
- Kerberoasting, DCSync attempts or Pass-the-Hash against domain controllers
- Unexpected outbound calls to LLM APIs (api.anthropic.com, api.deepseek.com, open.bigmodel.cn, api.x.ai)
- Traffic to webhook.site
- Traces of WISP, vshell or AdaptixC2
How to remediate
The order matters. Contain first, preserve evidence, then clean up. Deleting things early destroys what you need to know what happened.
Contain now
- Keep the affected ConnectWise app connections disabled until ConnectWise publishes formal restoration instructions.
- Don’t delete the enterprise applications as a first step. Disable them and keep their configuration as evidence.
Preserve evidence
- Export Entra sign-in and audit logs, Purview audit logs, SIEM data and any Graph activity logs you have. Don’t hold up containment while the exports run.
- Review from at least October 1, 2026, and look at earlier failed sign-ins and unusual volumes too.
Clean up what you confirm
- Have an authorized responder remove unauthorized credentials, consent grants, owners, role assignments and identities, after the evidence is saved.
- Rotate credentials and secrets the attackers could have reached, including Key Vault secrets, and revoke permissions that aren’t needed.
- Confirm ConnectWise’s own credential rotation took effect and no unauthorized access remains in your tenant.
- Check how disabling the apps affects your backups and restores, and follow ConnectWise’s current restoration guidance.
Harden for next time
- Give multi-tenant apps the least privilege they need, and review consent grants regularly.
- Alert on new app credentials, privilege assignments and unusual app-only activity, especially for high-privilege apps.
- Send Entra, Graph and Purview logs to a SIEM or Log Analytics with long retention, so the next investigation isn’t limited by default retention.
- Take inventory of everything you expose to the internet, including partner portals, legacy services and third-party integrations. Put admin interfaces behind trusted networks or Zero Trust access. ARTEX goes after exactly these overlooked systems.
- Scan code, pipelines and configs for embedded secrets, and rotate any you find.
If you find signs of compromise, the FBI asks you to report it to your local FBI field office or to IC3.
The bigger lesson
ARTEX makes attacks faster, not new. Every step in this campaign used weaknesses defenders already know about: trusted third-party apps with broad permissions, weak or reused credentials, and short log retention. AI has shrunk the time between “someone found a gap” and “someone used it” from weeks to hours. The fixes are the familiar ones, but they now need to be in place before the attack, because there won’t be time after.
Sources
- FBI FLASH-20261113-001, Threat Overview of ARTEX Autonomous Penetration Testing Framework, November 13, 2026 (TLP:CLEAR)
- Microsoft Security Blog: Storm-0940 ARTEX campaign and ConnectWise customer guidance, October 21, 2026
- BleepingComputer: FBI warns ARTEX AI hacking tool is powering global cyberattacks
- TechCrunch: FBI says Artex AI open source hacking tool used to breach networks in under three hours, November 16, 2026
- ConnectWise Trust Center advisories and Cloud Backup remediation guide