← cd ../blog

CVE-2026-84869: ScreenConnect clients run files pushed by guests without asking

A missing-authorization flaw in the ConnectWise ScreenConnect client (CVSS 9.9, CISA KEV) lets session guests push and run files on technicians' machines. Exposure checks, hunting and the fix.

Most ScreenConnect bugs we’ve had to patch over the years were on the server. CVE-2026-84869 is different: it’s in the client, and ConnectWise is explicit that “ScreenConnect servers are not impacted.” So patching the server and moving on leaves you exposed. The fix only lands when every installed client is on 26.6.5 or later. CISA added it to the KEV catalog on 11 September 2026, and it’s being used against MSP technicians. This is the second of three KEV entries I’m covering this week, after F5 BIG-IP APM.

Scope and ethics. This post covers architecture, exposure checks, detection and remediation for systems you own or are authorized to assess. It contains no exploit code and no step-by-step attack. The IOCs come from published research and are here for hunting.

Updated 20 Nov 2026 with Huntress’s technician-side exploitation findings.

At a glance

   
CVE CVE-2026-84869
Severity CVSS 3.1 9.9 Critical AV:N/AC:L/PR:L/UI:N/S:C/C:H/I:H/A:H (ConnectWise, via NVD)
CWE CWE-862 Missing Authorization, CWE-269 Improper Privilege Management
Affected ScreenConnect client (host client and access agents) before 26.6.5. Servers are not affected
Fixed in 26.6.5 (bulletin dated 8 Sep 2026). Cloud servers already updated; clients must be reinstalled
Exploited? Yes. CISA KEV, added 11 Sep 2026 (federal due date 14 Sep), forensic triage required. Huntress reports in-the-wild exploitation
Auth / interaction A participant in a session (low privileges) / no confirmation prompt on the target

Description: how the bug works

ScreenConnect (formerly ConnectWise Control) is a remote support and remote access tool. It has two kinds of participants in a session: the host (the technician driving the session) and the guest (the machine or person on the other end). Both run a ScreenConnect client, and the session supports features like file transfer and “run file.”

Those actions are supposed to be gated by role permissions (TransferFiles) and, in some cases, by a confirmation prompt on the receiving side. ConnectWise says the client’s file-transfer handling “did not properly enforce authorization or Host confirmation for file-transfer actions during an active session.” In certain circumstances, a file could be transferred to the host and executed there, “including through elevated execution actions.”

That flips the trust model of remote support. Usually the technician pushes tools to the customer. With this bug, a guest can push a file to the technician’s workstation and have the technician’s own client run it. ConnectWise hasn’t published the exact code path, citing the issue’s sensitivity.

Impact

  • Code execution on the technician’s workstation as the user running the client, with a path to elevation per ConnectWise.
  • MSP blast radius. A technician workstation usually holds the RMM console, PSA, password vault and customer admin credentials. Huntress lists all of these as credentials to rotate after a compromise.
  • Worm-like spread. Huntress and SecurityWeek describe chains where a compromised machine installs a rogue ScreenConnect client and propagates to the next session.

Which deployments are exposed

Component Status
ScreenConnect host client (technician side) before 26.6.5 Vulnerable. This is where Huntress saw exploitation
Access agents / installed clients before 26.6.5 Affected per ConnectWise, update them
ScreenConnect server (cloud or on-prem) Not affected, but it’s how you push updated clients
Cloud tenants Server updated by ConnectWise; you still need to reinstall host clients and update access agents

Prerequisites: an active session with an unpatched client. According to Huntress, the attacker doesn’t need a modified client or a stolen technician account. They need the technician to join a session where the attacker is a guest, either by joining a real customer session as an extra guest or by hosting one themselves. The rogue client shows up after execution, as a payload.

Identifying affected systems

On the server, check the version under Administration > Overview, along with “Latest Eligible Version” (on-prem must be on 25.4 or later and under maintenance to get 26.6.5). Then inventory the clients on endpoints you manage:

# Windows: find ScreenConnect client binaries and their versions (installed agents and per-user host clients)
$paths = "$env:ProgramFiles", "${env:ProgramFiles(x86)}", "C:\Users\*\AppData\Local"
Get-ChildItem -Path $paths -Recurse -Include ScreenConnect.ClientService.exe,ScreenConnect.WindowsClient.exe -ErrorAction SilentlyContinue |
  Select-Object FullName, @{n='Version';e={$_.VersionInfo.FileVersion}} |
  Sort-Object Version

# Installed ScreenConnect services (one per instance thumbprint)
Get-Service -DisplayName 'ScreenConnect Client*' | Select-Object Name, Status

Anything below 26.6.5 is affected. NVD’s CPE puts the boundary at build 26.6.5.9742. Remember to include the technicians’ own laptops, not only the customer fleet.

Safe validation in an isolated lab

You don’t need to reproduce the bug to check your posture:

  1. In a lab ScreenConnect instance, confirm your inventory script finds both an old and a 26.6.5 client, and that only the old one gets flagged.
  2. Apply the TransferFiles workaround to a test role and confirm a legitimate file transfer from that role is blocked.
  3. Run a benign “run file” (a script that writes the date to a text file) from a lab session and confirm your EDR records the ScreenConnect.WindowsClient.exe to cmd.exe lineage and that the server logs SentFiles/RanFiles events. If you can’t see benign activity, you won’t see malicious activity either.

Detection and threat hunting

Server audit data (Huntress). Work on a copy of App_Data\Session.db.

  • Event types SentFiles (EventType 86) and RanFiles (EventType 85).
  • ProcessType = 2 means a guest connection, ProcessType = 1 a host connection. Exposure is confirmed when a guest pushed a file while a host connection was open.
  • Red flags: ../ path traversal in file names, Startup-folder paths, high-volume RanFiles from one guest session, and blank-named guest sessions repeatedly requesting update*.bat. Huntress recorded about 688,000 such requests across 42 MSP servers between 20 Oct and 16 Nov 2026.

Endpoint process lineage.

  • ScreenConnect.WindowsClient.exe spawning cmd.exe, powershell.exe, wscript.exe or cscript.exe from a Temp path, with RunFile in the parent’s command line.
  • curl downloading an .msi, then msiexec.exe /a ... TARGETDIR=%LOCALAPPDATA%\..., then runas /trustlevel:0x20000.
  • seclogon service switched to automatic start.
  • New files in the user’s Startup folder (Huntress saw Sync.bat).

Published IOCs (Huntress) for hunting, not blocking alone:

  • Rogue client: %LOCALAPPDATA%\wclient_<8hex>\Windows Update Agent\WindowsUpdate.WindowsClient.exe, MSI at %TEMP%\wupdate_<8hex>.msi, SHA256 f8871b1561d5c77c297378a7f49c84de52a01ff297395544f0e95d13ef972841
  • Relay and payload host masterdarky[.]com (port 8041), resolving to 194.15.36[.]33
  • Guest-pushed files: update_<8hex>.bat, marker.txt, Run-Lateral.bat, svc_payload.cmd
  • Huntress also lists 27 sprayer IPs and several other guest IPs in its post

Network. Outbound ScreenConnect relay traffic (commonly port 8041) to a relay host you don’t own is a strong signal of a rogue client.

Huntress’s earlier write-up on rogue ScreenConnect installations (late August) describes a related VBScript chain (1.vbs–4.vbs, WindowsServiceHost.vbs, XMRig as SearchIndex.exe). They recommend reimaging any device with script execution from Process: Guest in the audit log.

Remediation

  1. Update every client to 26.6.5 or later: technician host clients and access agents. On cloud, ConnectWise updated the server; you must still reinstall host clients and update agents. On-prem, upgrade the server to 26.6.5 first, then roll out clients.
  2. Hunt the server audit data and technician endpoints using the indicators above, especially for sessions since late August.
  3. If a workstation was compromised: isolate it, remove the rogue client and its MSI, reset seclogon to its normal start type, reimage, and rotate every credential reachable from it: RMM, PSA, ScreenConnect, password vault and customer admin accounts.
  4. After patching, ConnectWise recommends reviewing users and removing unknown accounts, reviewing roles and permissions, changing passwords and enabling MFA.

Interim mitigation (if you can’t patch today)

  • ConnectWise’s workaround: in Administration > Security > Roles, edit each role and deselect TransferFiles for each session group, then save. This reduces exposure but isn’t a fix.
  • Restrict guest permissions, limit access to the web interface and installers by IP or VPN, and don’t join unattended sessions you can’t tie to a real customer (Huntress).
  • Delete spray sessions (blank-named guests requesting update*.bat) and keep checking, since Huntress says they come back.

References

Verify every version against the vendor advisory before acting. Published 05 Nov 2026.