Two separate campaigns this week turned ordinary remote access software into backdoors: a Windows attack run through ScreenConnect panels and a growing wave of GSocket infections on Linux web servers.
Published: | Last reviewed:
Attackers are installing genuine remote access software as a backdoor, and patching the original hole does not remove it. On November 11, 2026, Huntress reported rogue ScreenConnect clients on at least 288 endpoints across 48 organizations after attackers ran commands from cloud-hosted ScreenConnect panels. Separately, Microsoft reported that devices with GSocket detections on Linux servers rose 356.2% between March and August 2026. No source opened for this article links the two campaigns.
Key takeaways
- Huntress counted 288 endpoints in 48 organizations hit within about 90 minutes of 17:47 UTC on November 11, and the count covers Huntress customers only.
- ConnectWise says it has no evidence of a ScreenConnect flaw or a breach of its own systems, and it disabled five third-party extensions in about 400 partner environments.
- Huntress has not seen the worm’s technician-to-technician spread succeed. The takeover step failed in every recovered host log it checked.
- Do not investigate a suspect host through ScreenConnect. Use EDR live response or another out-of-band tool.
- Microsoft’s GSocket figures cover March to August 2026, and web applications and hosting made up 87.1% of attributable incidents.
- GSocket runs through a shared public relay, so there is no attacker IP address or domain to block. Egress control is the lever.
- Patching a compromised web application leaves a running GSocket client and its persistence in place.
Rogue remote access client: a legitimate remote administration agent, such as a ScreenConnect client or GSocket’s gs-netcat, that an intruder installs to keep control of a machine. Because the software is genuine, the intruder’s access no longer depends on the original entry point, and it keeps working after that entry point is fixed.
Are the ScreenConnect and GSocket attacks part of the same campaign?
The ScreenConnect and GSocket attacks are not linked by any source opened for this article. Huntress tracked SCKit, a Windows toolkit pushed through hijacked ScreenConnect panels, while Microsoft described GSocket abuse by several unrelated threat actors on Linux web servers. What they share is the method: a legitimate remote access tool becomes a backdoor that keeps working after the first foothold is closed.
Huntress reported on November 11, 2026 that an operator used access to several cloud-hosted ScreenConnect panels to push commands to groups of managed endpoints. The commands installed attacker-controlled ScreenConnect clients and a toolkit called SCKit on at least 288 hosts across 48 organizations that use Huntress. The activity began around 17:47 UTC, and every infection Huntress counted landed within 90 minutes [1].
ConnectWise said it found no evidence of a compromise of its own infrastructure or a flaw in ScreenConnect, and it disabled five third-party extensions in about 400 partner environments as a precaution [2] [4].
Microsoft published a separate report the same day. It describes unrelated threat actors who deployed GSocket on compromised Linux servers and kept access after the original vulnerabilities were patched, and its Defender for Endpoint telemetry showed a 356.2% rise in devices with GSocket detections from March to August 2026 [3].
How did attackers push rogue ScreenConnect clients to 288 endpoints?
The rogue ScreenConnect clients were pushed through the Run Command feature of compromised cloud-hosted panels, which let one operator send a command to many endpoints at once, running as SYSTEM. Each command installed an attacker-controlled client silently, then dropped a Base64-encoded batch file that fetched the SCKit toolkit. Huntress counted all 288 infections inside 90 minutes.
A ScreenConnect panel can push commands that run as SYSTEM to many managed endpoints at once [1].
About one third of the hosts were hit in the first 10 minutes, according to Huntress [1]. That pace points to a scripted bulk push from the panel, which is an inference from the timing and not a Huntress finding.
The install sequence on each endpoint, as Huntress reconstructed it [1]:
- The command checks for an existing ScreenConnect client, downloads an attacker-controlled installer to
C:\Windows\TEMP\sc.msiand installs it silently. - A Base64-encoded dropper,
sckit-run.bat, lands inC:\Windows\Tempand pulls down a decoder script. - The decoder requests a toolkit keyed to that machine. The server returns decryption values, and the script unpacks 16 files into the temp folder.
- The kit moves to
C:\ProgramData\RelayTrap, adds Microsoft Defender exclusions and installs a second rogue client. - A scheduled task named TrapWatch runs as SYSTEM every 10 seconds, stops the legitimate ScreenConnect service and positions the attacker’s copy to replace it.
Is ScreenConnect itself vulnerable, and how did the attackers get in?
ScreenConnect itself is not reported as vulnerable in this campaign. ConnectWise says it has no evidence of a flaw in the software or a compromise of its infrastructure, and it suspects abuse of independent third-party extensions. Huntress has seen no evidence of a new vulnerability and points to compromised credentials or hijacked operator sessions, and the initial access method remains unconfirmed.
ConnectWise posted its advisory at 22:45 UTC on November 11 and named five extensions: Labtech Plus, RMM Plus, RMM+, RMM Plus Password Link and RMM+ Password Link [2]. The company says they expose an API able to run remote commands, protected by a passphrase the partner sets during setup, and that their presence alone does not establish impact [4].
Huntress told The Hacker News that its telemetry currently points to compromised credentials or hijacked operator sessions, and that the route to panel access is unclear [4]. ConnectWise lists cause and scope as under investigation, so the two explanations sit side by side for now [2].
One detail narrows things down. The affected builds are 26.6.6.9747 and 26.7.6.9783, and ConnectWise’s earlier file-transfer fix shipped in 26.6.5 or higher [1] [2]. Both builds are newer than that fix, so the activity does not look like CVE-2026-84869, though that is my reading of the version numbers and not a vendor statement.
The ScreenConnect campaign rests on claims of very different strength, so the table below separates what each source observed, stated or only suspects.
| Claim | Evidence status | Source |
|---|---|---|
| Rogue clients on 288 endpoints in 48 organizations | Observed by Huntress, Huntress customers only | [1] |
| Five third-party extensions disabled | Action taken by ConnectWise | [2] |
| Extensions were the entry point | Suspected by ConnectWise, not confirmed | [2] [4] |
| Compromised credentials or hijacked sessions | Suggested by Huntress telemetry, not confirmed | [4] |
| Spread to technician workstations | Designed in the code, not observed to succeed | [1] |
| New ScreenConnect vulnerability | No evidence reported by either vendor | [4] |
UnpanicTech did not reproduce either campaign or test any detection. Everything here comes from the vendor reports listed at the end.
Did the SCKit worm really spread from technician to technician?
The SCKit worm has not been seen spreading from a technician workstation to other endpoints. Huntress reconstructed that capability from the toolkit’s code and configuration, and the console takeover attempt failed in every recovered host log it checked. What Huntress did observe was operators pushing the toolkit through panels, so the spreading behavior exists only as a designed capability so far.
Huntress describes a two-stage design. An infected endpoint waits for a technician to connect, then pulls a dropper and loader that end in a script stealing the technician’s .aspxauth session cookie and replaying it against the panel [1]. From there the code is built to upload a payload to the panel’s Shared Toolbox, tell every reachable session to run it and delete the item about 60 seconds later [1].
The unproven part still matters for response planning. The intended bridge is a technician who connects to an infected host through ScreenConnect, which is why Huntress tells responders to use EDR live response, WinRM or another out-of-band tool instead [1]. The EDR vs XDR vs MDR comparison shows how those options differ.
If your incident response plan assumes you can remote into a suspect machine with your usual tool, this is a good week to rewrite that line.
Huntress also noted that the RelayTrap source appears to have been written with an LLM, judging by its detailed comments [1]. That is a comment on code style, and it says nothing about who is behind the activity. For the wider trend, see our piece on autonomous AI attacks and the six months defenders have.
Why are GSocket attacks on Linux servers growing?
GSocket attacks on Linux servers are growing because the tool removes problems attackers usually face after breaking into a web application. It needs no attacker-owned server, it connects outward through a shared public relay, and its deployment script handles hiding and persistence. Microsoft reports a 356.2% rise in devices with GSocket detections between March and August 2026, though the increase was uneven month to month.
GSocket is an open source remote access tool from The Hacker’s Choice, and Microsoft cites its GitHub project in the report [3]. Microsoft explains the adoption by comparing it with older options: SSH keys are easy to spot in an audit, reverse shells depend on servers defenders can block, and custom implants take effort to maintain [3].
The GSocket telemetry from Microsoft breaks down like this:
| Measure | Figure | Context |
|---|---|---|
| Growth in devices with GSocket detections | 356.2% | March to August 2026, uneven by month |
| Web applications and hosting | 87.1% | Share of attributable incidents |
| PHP web apps and cPanel hosting | 63.9% | Share of that web and hosting group |
| Apache | 24.3% | Share of that web and hosting group |
| Education and research | 30.7% | Share of affected organizations |
| Technology and telecommunications | 22.9% | Share of affected organizations |
Source: Microsoft [3], with the web and hosting split also reported by BleepingComputer [5].
GSocket gives defenders no attacker-owned IP address or domain to block, because the relay is shared public infrastructure [3].
Microsoft cautions that the sector shares reflect exposure and do not show that attackers chose those sectors. Universities often run large, decentralized web estates that are hard to inventory and patch [3]. The numbers also come from Defender for Endpoint, so they describe Microsoft’s customers and not every Linux server on the internet.
How does GSocket keep working after the web app is patched?
GSocket keeps working after a patch because it runs as its own process, separate from the exploited web application. The compromised server and the operator both connect outward to the Global Socket Relay Network, which pairs the two sessions over an encrypted channel. A shared secret, not an IP address, identifies the server, so the operator can reconnect later from anywhere.
Microsoft’s analysis describes a repeatable sequence observed across several remote code execution intrusions. Attackers exploit an internet-facing application, then use curl to download and run a script that profiles the host, picks a GSocket client for its architecture and stages it in a writable or memory-backed path such as /dev/shm [3]. The client then opens an encrypted outbound connection to the relay network, which Microsoft describes as third-party infrastructure the attackers neither own nor host [3].
To stay hidden, the attackers renamed the process to look like a kernel thread, set up persistence through cron, systemd, shell profiles and watchdogs, and deleted staging files [3]. Because the server is identified by a shared secret and not an address, blocking an IP or domain is unlikely to cut off access. In a few incidents the attackers also tried the Dirty Frag kernel flaw, CVE-2026-43284, to reach root [3].
A patched web application can still host a live backdoor [3].
Microsoft adds two limits to its own findings. Not every stage occurred in every intrusion, and persistence across reboot and a successful relay session were not confirmed in every case [3]. Treat the attack chain as a generalization, not a script every attacker followed.
What should defenders check on Windows and Linux hosts today?
Defenders should stop trusting the remote access tool itself and check for what it left behind. On Windows, audit ScreenConnect Run Command and Toolbox history, look for a second ScreenConnect client, and investigate suspect hosts without connecting through ScreenConnect. On Linux, remove the GSocket client, its persistence and staged files, confirm outbound relay connections have stopped, and restrict egress to approved destinations.
What should ScreenConnect administrators do first?
ScreenConnect administrators should start with the panel, because that is where the commands came from. Hunting for this kind of legitimate-tool abuse overlaps with hunting living-off-the-land activity in EDR telemetry.
- Review Run Command and Toolbox history for mass activity you did not start, and look for an unauthorized second ScreenConnect client on managed endpoints [1].
- Confirm the five extensions are disabled. On-premises partners must do it themselves because ConnectWise cannot act on their behalf [2].
- Do not connect to a suspect host through ScreenConnect, and treat any technician who already did as potentially exposed [1].
- Keep isolated hosts isolated until cleanup finishes, because Huntress saw systems reinfect shortly after isolation was lifted [1].
- From a clean device, end active ScreenConnect web sessions, rotate affected credentials and revoke exposed browser and identity provider sessions, since the toolkit can steal session tokens [1]. Microsoft accounts need their own audit, which the guide to investigating compromised M365 and Entra ID accounts walks through.
What should Linux server owners do first?
Linux server owners should treat a patched web application as the start of the cleanup, not the end. A Linux server hardening baseline covers several of these items already.
- Remove the GSocket client, its persistence and staged files, then verify that outbound connections to the relay have stopped [3].
- Apply deny-by-default egress rules and allow only destinations the server needs. The network segmentation best practices guide explains how to scope that.
- Mount temporary and memory-backed directories as non-executable where feasible, and alert on execution from hidden or world-writable paths [3].
- Audit cron entries, systemd units with plausible service names and shell startup files, since persistence is often layered [3].
- Patch local privilege escalation flaws on internet-facing servers, including CVE-2026-43284 [3] [5]. A solid vulnerability management process keeps that from slipping.
Both campaigns end the same way for responders: removing the original entry point is only half the job, and the other half is finding what the attackers left running. For GSocket, Microsoft says defenders should remove the relay client, its persistence and its staged files, then verify that outbound connections to the relay have stopped, because patching the application does not remove the client [3]. For the ScreenConnect activity, Huntress says cleanup has to remove the rogue client, the TrapWatch task and the RelayTrap folder before a machine returns to service, because isolation alone does not remove the persistence [1].
Evidence matters too. ConnectWise asks partners to preserve logs of suspicious activity and to watch for unexpected remote commands, Toolbox activity and unapproved remote access software on customer devices and technician workstations [4]. Microsoft pairs its Linux cleanup advice with persistence audits and deny-by-default egress rules [3].
The host artifacts below come straight from the vendor reports, and they are defensive search terms only. Huntress warns that file hashes are weak indicators for SCKit, because the scripts are rebuilt and the trap is compiled separately on every host [1].
| Platform | Artifact to search for | What it indicates | Source |
|---|---|---|---|
| Windows | Scheduled task TrapWatch | Task that swaps in the attacker’s client | [1] |
| Windows | C:\ProgramData\RelayTrap | Folder holding the SCKit trap | [1] |
| Windows | Run key value RelayTrapLiteAlert | Startup persistence for the trap | [1] |
| Linux | /dev/shm/.gs-0 | Hidden staging directory for gs-netcat | [3] |
| Linux | systemd-remount-d, defunct, defunct.service | Service-like names used to disguise the client | [3] |
| Linux | /tmp/.gsn_ora | Hidden file flagged as a GsNetcat tool | [3] |
What is still unknown about both campaigns?
Several basic facts about the ScreenConnect campaign and the GSocket activity remain unconfirmed. Nobody has confirmed how attackers reached the ScreenConnect panels, how many managed service providers are affected, or who is responsible. For GSocket, Microsoft says persistence across reboot and a successful relay session were not confirmed in every incident, so the reported chain is a generalization.
- The ScreenConnect entry point. ConnectWise lists the initial access method and full scope as under investigation [2].
- The real victim count. Huntress’s 288 covers its own customers, and the sequential tags it saw on compromised providers start partway through the sequence, which suggests earlier victims [4].
- The actor. None of the sources opened for this article names one.
- The technician-to-panel spread. It is designed in the code and unproven in the field [1].
What are the strategic implications for MSPs and Linux server owners?
Rogue remote access clients change what “fixed” means for MSPs and Linux server owners. Cleaning a ScreenConnect panel or patching a web application does not remove a client that already runs on an endpoint. Remediation needs its own checklist, separate from patching, covering host isolation, persistence removal and outbound connection monitoring.
Huntress saw hosts reinfect once isolation ended [1], and Microsoft says patching the application leaves the GSocket client untouched [3]. For MSPs, the lesson is blast radius: one ScreenConnect panel can push SYSTEM commands to many endpoints, so operator credentials, session tokens and any third-party extension with a remote command API deserve the same scrutiny as the endpoints themselves (my reading of the ConnectWise advisory). Operator accounts should use phishing-resistant MFA with conditional access, and the cookie theft in SCKit shows why revoking sessions matters as much as rotating passwords.
For Linux server owners, GSocket moves the defense from blocking attacker infrastructure to controlling what a server may reach. Rogue clients that depend on a shared relay are better caught by egress allowlists, non-executable mounts and persistence audits than by IP blocklists, and the zero trust microsegmentation versus legacy firewall comparison covers the outbound side. A practical first step this week: list every ScreenConnect extension you run and every Linux server with unrestricted outbound access.
Sources & References
- Huntress, “SCKit Worm Installs Rogue ScreenConnect Clients and Session Cookie Relays Across Hundreds of Endpoints,” November 11, 2026
- ConnectWise Trust Center, Latest Advisories (November 11, 2026 ScreenConnect security advisory; September 3, 2026 file-transfer advisory)
- Microsoft Security Blog, “The global socket: how threat actors are weaponizing GSocket for stealthy C2,” November 11, 2026
- The Hacker News, “ScreenConnect Attacks Target at Least 288 Endpoints Across 48 Organizations With Rogue Clients,” November 12, 2026
- BleepingComputer, “Hackers plant GSocket on Linux servers for persistent access,” November 12, 2026
Investigating unauthorized RMM binaries or encrypted gsocket relays? Make sure you defang IOCs before logging relay domains and deploy the hunting queries in our Sigma and YARA rules repository.
- Host Forensics & Telemetry: Run live socket and process checks using our Linux forensics cheat sheet and correlate Windows RMM installs against Sysmon event IDs.
- Lock Down Administrative Access: Prevent unauthorized remote access tunnels with Linux bastion host hardening and our full incident response guide.

Leave a Comment (Authenticated Users)