A guest in a ScreenConnect session made an MSP technician’s own client run a batch file on November 12. Within roughly nine seconds, that workstation had launched a rogue remote-access client pointing to an attacker-controlled relay. Huntress traced the execution to exploitation of CVE-2026-84869, a client-side file-transfer authorization flaw. The technician needed only to join the session; the client showed no confirmation prompt.

That puts the workstation used to administer customer systems at risk. Huntress also saw a second technician’s client open a guest-pushed test file on November 13. Its search across 42 MSP servers found 688,000 guest “run file” requests from 29 source IP addresses since October 20. The count measures guest requests; check technician workstations for execution before counting affected hosts. ConnectWise published the vulnerability in September, and NVD records its September 11 addition to CISA’s exploited-vulnerability catalog. This November reporting adds a completed workstation intrusion and server-side records that defenders can match to technician connections. Nulltap’s earlier ScreenConnect vulnerability notice covers the disclosure and initial fix.

A guest file reached the host

In Support and Access sessions, a guest can ask the technician’s ScreenConnect host client to transfer and execute a file without authorization or host confirmation. ConnectWise’s CVE record describes the affected component as the client and rates the flaw 9.9 on CVSS v3.1. The attacker needs access as a session guest, either through a machine they control or as another participant in a real customer support session.

The November 12 process trace starts with the host client launching update_224ac089.bat from its Documents\…\Temp\ area. The file relaunched hidden, sent a notification through a Telegram bot, checked the Windows build and admin rights, then downloaded an MSI from masterdarky[.]com. Huntress observed the installer unpacked as an administrative image into %LOCALAPPDATA%\wclient_224ac089, leaving no normal install record. The resulting WindowsUpdate.WindowsClient.exe connected to the attacker’s relay on port 8041. That name mimicked a Windows component, but Huntress identified it as a rebranded ScreenConnect client.

The operator had already sent requests for that specific batch filename to five MSP servers, beginning 43 minutes before the workstation run. Each server received the request hundreds or thousands of times. Huntress assessed with high confidence that the spray and the masterdarky[.]com operator were the same. Huntress described the repeated requests as “waiting for any technician to join.” The report also describes separate SCKit server-side activity and crypto-named guest sessions; treat those as separate investigations rather than counting them as confirmed use of this client flaw.

Session overlap separates attempts from exposure

The server’s App_Data\Session.db records guest-originated RanFiles events (EventType 85) and SentFiles events (86). Huntress identifies a guest connection by ProcessType 2 and a technician host connection by ProcessType 1. Work from a copy of the database. Pair each guest request with the host connection window for that same session. A matching file request while a host is connected is an exposure to investigate on that workstation; a request outside that window is an attempt. Prioritize overlapping requests for workstation process checks.

On October 19, seven pushes of Run-Lateral.bat missed a vulnerable technician client; the closest arrived 4.3 seconds after disconnect. On November 13, a different guest at 91.224.92[.]167 sent marker.txt after a technician joined, and the client opened it in Notepad++. That observed run establishes working guest-to-host execution on that machine, while the text file itself was a test. In October, attacker guests also requested a path ending in STARTM~1/Programs/Startup/Sync.bat during three real customer support sessions. Two connected technicians used vulnerable Windows clients; the third used a Mac, where that Windows Startup path failed. Investigate the two Windows hosts for a dropped script and any Startup-file execution.

Rebuild technician exposure, then patch

First, inspect server connection history for the host-client builds technicians actually used. ConnectWise’s CNA record marks versions before 26.6.5 affected. Bring on-prem host clients and Access agents to 26.6.5 or later. For ConnectWise cloud deployments, reinstall each host client and bring Access agents up to date as the vendor recommends. Automate-integrated on-prem deployments with active Automate Assurance can apply the 26.6.5 product update through Automate. Updating only the server leaves an older technician client in the execution path. For a delayed rollout, use Administration > Security > Roles to remove TransferFiles from assigned session groups for each role, save, and schedule the client update. ConnectWise calls that a temporary mitigation.

Next, query a copy of App_Data\Session.db for guest RanFiles and SentFiles, especially repeated update.bat or update_<8hex>.bat names, marker.txt, Run-Lateral.bat, svc_payload.cmd, and traversal segments such as ../. Match the guest event to the technician host’s session time and installed build. On the overlapping workstation, look for the branded or standard ScreenConnect.WindowsClient.exe spawning .bat, .cmd, .ps1, .vbs, .js, or .txt from its Documents\*\Temp\ area. Then check %APPDATA%\Microsoft\Windows\Start Menu\Programs\Startup\ for unexpected files, %TEMP%\wupdate_<8hex>.msi, %LOCALAPPDATA%\wclient_<8hex>, and WindowsUpdate.WindowsClient.exe. Inspect outbound relay destinations, particularly port 8041, against your own ScreenConnect relay inventory. The installer’s administrative-image extraction means an empty installed-software listing is a poor endpoint check.

Huntress names the following IPs for the spray: 162.35.40[.]249, 64.20.44[.]120, 103.195.102[.]66, 103.195.102[.]76, 199.127.60[.]150, 199.127.62[.]58, 181.215.242[.]106, 162.35.167[.]105, 23.92.65[.]131, 23.94.77[.]110, 198.23.165[.]181, 107.173.186[.]58, 50.114.203[.]14, 38.18.228[.]165, 66.118.239[.]23, 74.235.98[.]176, 172.81.61[.]149, 178.173.246[.]85, 154.30.4[.]248, 107.148.223[.]103, 103.195.100[.]70, 103.195.100[.]69, 38.18.229[.]249, 98.142.247[.]243, 45.13.233[.]32, 38.18.228[.]173, and 199.127.60[.]149. The report also names 194.15.36[.]33 as the relay-domain resolution, 91.224.92[.]83 for the October near-miss, 193.26.115[.]63 and 57.128.131[.]66 for Startup-path requests, and 107.148.223[.]37 for svc_payload.cmd. Correlate those IPs with session and process evidence instead of treating a network match alone as proof of execution.

Restore trusted technician access

If the workstation ran a guest file or contains the rogue client, isolate it and investigate the administrator access reachable from it. Remove the unauthorized client and its staged MSI, return the seclogon service to your documented baseline, and rotate reachable RMM, PSA, ScreenConnect, vault and customer-admin credentials. Restrict guest joins to authorized support sessions and limit access to the ScreenConnect web interface and installers by VPN or IP where possible.

For verification, confirm that each active technician host client reports at least 26.6.5, then replay the server’s guest-file events against host-connection windows and endpoint process histories. A patched build fixes the known client flaw; a historical overlap plus client-launched file is incident evidence that calls for workstation response.