Ransomware, IR & Endpoint

SCKit and the Crypto-Bait Sessions: Two Routes Into a ScreenConnect Panel

Dark cyberpunk illustration of a vast grid of glowing cyan glass cubes receding into darkness, where two cubes in the middle distance are hollow amber outlines holding nothing and one cube at the front burns amber from inside.

One hundred and nine companies have machines in their remote-support console that do not exist. The session names pair a crypto exchange with a role word in either order - KUCOIN-SRV, LAPTOP-BINANCE, BACKUP-PROBIT, WORKSTATION-COINEX - and each one shows a trading-desktop thumbnail and reports as online around the clock. The first three were enrolled on November 3 at 21:26 UTC from a residential address in Egypt. By November 13, Huntress had counted somewhere between 2,223 and 2,408 of these sessions across 118 self-hosted ConnectWise ScreenConnect servers, using more than 1,000 distinct fake names.

Two days earlier a separate operator ran a separate campaign against the same product. SCKit reached 288 hosts across 48 organizations inside 90 minutes on November 11, about a third of them in the first ten, then came back on November 13 for 205 hosts across 41 organizations. Huntress found no indicator overlap between the two campaigns. Two unconnected crews, one week, one remote-support product, and in both cases the target was the console and the technician in front of it rather than the server.

Here is who needs to act. If you run a self-hosted ScreenConnect server, or your managed service provider does, the panel audit at the end of this article is today's work. If any ScreenConnect host client or Access agent in your estate is older than 26.6.5, you are inside the window both campaigns were built around. If there is no ConnectWise ScreenConnect anywhere in your environment or your providers', the file paths and addresses below are not yours, and you can skip to the last section - the shape of this applies to whatever remote monitoring tool you do run.

The console deserves more attention than the server because of what it already holds. A ScreenConnect panel carries a pre-authorized, SYSTEM-level execution path to every endpoint enrolled in it. An operator who can issue commands from that panel needs no exploit, no credential stuffing and no payload that evades anything. SCKit proved the point by using the product's own Run Command feature, and the panels it ran on were current: Huntress recorded ScreenConnect 26.6.6.9747 and 26.7.6.9783, both newer than the version that fixed September's critical bug.

Server upgrades leave the clients behind

The September flaw is CVE-2026-84869. The NVD record scores it 9.9 critical, vector CVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:C/C:H/I:H/A:H, and its description ends with a sentence most summaries drop: "ScreenConnect servers are not impacted." The defect lives in the client. A low-privilege authenticated user can transfer and execute files through an active session without host authorization or confirmation. CISA added it to the Known Exploited Vulnerabilities catalog on September 11 with a due date of September 14, a three-day window under BOD 26-04. ConnectWise shipped the fix in 26.6.5 on September 8.

Server upgrades do not carry that fix to the technician laptops or the managed endpoints. Those are separate binaries with separate version numbers, and in most shops nobody owns them. ConnectWise's advisory on the night of November 11 asked partners to verify that every technician client runs 26.6.5 or later. A vendor repeating that two months after a KEV listing marks the gap both of this week's campaigns were built to sit in.

Three version numbers, not one

  • The server build. Cloud instances were updated by ConnectWise. On-premises instances need 26.6.5 or later, and the server history page will tell you what it is running.
  • The host client on every workstation a technician uses to join sessions. This is the component CVE-2026-84869 actually affects, and the one nobody patches on a schedule.
  • The Access agent on every managed endpoint. Older agents persist through server upgrades.

Alongside the client advisory, ConnectWise identified five third-party extensions as possibly related to the SCKit activity and disabled them across more than 400 partner environments, including RMM Plus, RMM+, Labtech Plus and two Password Link variants. Huntress found that RMM+ versions 1.0.31 and 1.0.32 accepted a blank access key when none had been configured, because the RMMPlus.AccessKey setting shipped with an empty default. RMM+ bypasses ScreenConnect's own user accounts, roles and multi-factor enforcement, so a blank key is a full-privilege side door past every control you configured in the panel. Most installs Huntress saw were legitimate, which is exactly why nobody audits them.

SCKit asked the panel to do the work

The operator used Run Command on cloud-hosted panels to execute as SYSTEM across many endpoints at once. The command pulled an attacker-controlled installer to C:\Windows\TEMP\sc.msi and installed it silently, then wrote a base64 dropper to C:\Windows\Temp\sckit-run.bat, which fetched sckit-decoder.ps1. The payload request carries the machine's UUID and the server returns a key, IV and salt specific to that host, so file hashes are close to worthless here.

Sixteen files land. A loader decrypts a machine-specific k<32 hex>.ps1, which extracts the toolkit to C:\ProgramData\RelayTrap. From there install.ps1 registers a scheduled task named TrapWatch and a RelayTrapLiteAlert value under the Run key, and a second script adds Defender exclusions before installing a second attacker-controlled ScreenConnect client. TrapWatch.ps1 then runs as SYSTEM every ten seconds, stopping the legitimate ScreenConnect client, setting its service to manual start and taking over its connection slot. The replacement reuses the real client's libraries and connection details, so the endpoint still looks normal in the console it was stolen from. In one incident the rogue service was registered as Windows Update Agent, running UpdateConnect.ClientService.exe as LocalSystem out of C:\ProgramData\Windows Update Agent (de362a425905bcc4)\.

The detail I keep coming back to is the SRP and AppLocker rules the kit installs. They block files pushed through ScreenConnect's temporary folders. The operator closes the door behind itself, and the channel it closes is the one your technician would use to respond. That single design choice tells you how to handle these hosts: not over the tool that was compromised.

Huntress reconstructed the lateral path from code and configuration rather than observing it succeed, and labels it that way. About 2.5 seconds after a technician connects, a file transfer retrieves Run-Lateral-Full.bat; a loader pulls a per-machine kit; a script steals the technician's .aspxauth session cookie and replays it against the panel to enumerate reachable sessions; ToolboxRelayUploader.exe uploads a script to the panel's Shared Toolbox, broadcasts it to those sessions and deletes the item about 60 seconds later. The console-takeover script failed in every host log Huntress examined. The November 13 wave shows how the counting worked: installer requests carried a query string ending c=Exploited<N>, and one value, Exploited809, corresponded to 190 downstream clients from a single panel. Victims also had a netsh interface portproxy rule redirecting 127.0.0.1:88 to 185.241.211[.]197:80, with a modified hosts file to match.

The crypto-named sessions are bait for a technician

The second campaign does nothing at all, and that is its tell. Across 118 servers Huntress observed no file pushes, no commands and no guest-side activity. The sessions last about 64 minutes and reconnect roughly every 55, which keeps them green in the console. The operator revisits on a 151-second cycle and, over ten days, walked the same 100-plus servers in the same order every day. Server-side, every fake reports guest version 25.3.4.9288, ClientType 1, and an empty guest public key, and 97.8 percent of the connections came from one IONOS address, 74.208.158[.]106.

The leading hypothesis is that the fakes are bait, and the payoff is CVE-2026-84869 firing on the workstation of whichever technician gets curious and joins one. Huntress has not observed that exploitation and says so, and also cannot rule out simple keystroke capture; one technician used "send text" twice inside a fake session, which is a credential exposure on its own. Treat the mechanism as unproven and the exposure as real. The write-up itself gives two counts, 2,408 in the summary and 2,223 in the body, and does not reconcile them, so use the range.

One timeline item matters more than the indicators. In mid-October, weeks before any fake appeared, an online file-scanning service retrieved the victim MSP's publicly downloadable Access installer. A public installer link means anyone can enroll a machine into your console. Enrollment was never the attacker's hard problem.

What to run on the endpoints and on the panel

Three checks, in the order I would run them. First, the endpoint sweep for a rogue client and the trap. Run it from WinRM or your EDR's response shell, not through ScreenConnect.

# Run as SYSTEM/admin on each managed Windows endpoint. Any output is a finding.
$hits = @()
foreach ($p in @(
  'C:\ProgramData\RelayTrap',
  'C:\Windows\Temp\sckit-run.bat',
  'C:\Windows\Temp\sckit-decoder.ps1',
  'C:\Windows\TEMP\sc.msi')) {
  if (Test-Path -LiteralPath $p) { $hits += "PATH $p" }
}
if (Get-ScheduledTask -TaskName 'TrapWatch' -ErrorAction SilentlyContinue) { $hits += 'TASK TrapWatch' }
$run = 'HKLM:\SOFTWARE\Microsoft\Windows\CurrentVersion\Run'
if ((Get-Item $run).GetValue('RelayTrapLiteAlert')) { $hits += 'RUNKEY RelayTrapLiteAlert' }

# A second ScreenConnect-family service, or one under ProgramData, is the rogue client.
Get-CimInstance Win32_Service |
  Where-Object { $_.PathName -match 'ClientService\.exe|ScreenConnect\.ClientService' } |
  ForEach-Object { $hits += "SVC $($_.Name) :: $($_.PathName)" }

# Defender exclusions and the port-proxy rule the November 13 wave left behind.
(Get-MpPreference).ExclusionPath | ForEach-Object { $hits += "MPEXCL $_" }
netsh interface portproxy show all | Select-String '185\.241\.211\.197|:88'
$hits

Second, the technician workstations. This is the host-side pattern for CVE-2026-84869 exploitation: the ScreenConnect host client spawning a shell or an installer against a file in the user's session temp directory.

# On every workstation that joins sessions. Sysmon process creation, last 30 days.
Get-WinEvent -FilterHashtable @{
    LogName   = 'Microsoft-Windows-Sysmon/Operational'
    Id        = 1
    StartTime = (Get-Date).AddDays(-30)
} | Where-Object {
    $_.Message -match 'ParentImage:.*ScreenConnect\.WindowsClient\.exe' -and
    $_.Message -match 'Image:.*\\(cmd|powershell|wscript|cscript|msiexec)\.exe'
} | ForEach-Object {
    [pscustomobject]@{
        Time    = $_.TimeCreated
        Child   = ([regex]'\nImage: (.+)').Match($_.Message).Groups[1].Value
        Command = ([regex]'CommandLine: (.+)').Match($_.Message).Groups[1].Value
    }
} | Where-Object { $_.Command -match '\\Documents\\.*\\Temp\\' } | Format-List

Third, the panel itself. Preserve the session database and its write-ahead log before you delete anything, because the enrollment history is the only record of who joined a fake session.

:: On the ScreenConnect server, elevated. Copy first, query the copy.
mkdir C:\ir 2>nul
copy "C:\Program Files (x86)\ScreenConnect\App_Data\Session.db*" C:\ir\
sqlite3 C:\ir\Session.db ".dump" > C:\ir\session-dump.sql

:: Crypto-exchange session names, the emulator's guest build, and empty guest keys.
findstr /i "BINANCE KUCOIN PROBIT COINEX BITMEX OKX GATEIO" C:\ir\session-dump.sql
findstr /i "25.3.4.9288" C:\ir\session-dump.sql
findstr /i "74.208.158.106 185.241.211.197 185.99.255.28" C:\ir\session-dump.sql

If any of this lands, isolate the host and keep it isolated. Huntress reports reinfection where isolation was lifted early. Terminate every active ScreenConnect web session, rotate technician credentials from a clean device, and revoke browser and identity-provider sessions, because the kit reads browser cookie databases with a bundled sqlite3.exe. Then read the Run Command and Toolbox history in the instance audit log and account for every entry.

Account for every machine in your console

Open your panel today and reconcile it against your asset list. Every session should map to a customer, a site and a person who can confirm the machine exists. Anything with a blank company field, an unfamiliar name or a client version that does not match your fleet gets grouped out of the technician view and investigated. Make "do not join a session you cannot attribute" a written rule, bar triage over the tool itself, and put the host client and Access agent versions on the same patch report as the server. If your provider operates the panel, ask for all three version numbers in writing and when they last reconciled sessions against their asset list.

If the answer to any of that is uncomfortable, the work is an incident response plan that assumes the remote-support channel is the compromised one. That is a specific engagement, and it is the one we would run here.

Need an incident response plan before the next attack?

We help organizations build and test incident response playbooks, including the case where the remote-support channel itself is the compromised one. Book a session with our team.