Home/Blog/ScreenConnect MSP guest attack: what to ask your provider
blog

ScreenConnect MSP guest attack: what to ask your provider

November 18, 2026 · Security News

ScreenConnect MSP guest attack: technician workstation compromised via CVE-2026-84869 guest file run

Attackers sent about 688,000 “run file” requests through guest sessions on 42 MSP ScreenConnect servers between October 20 and November 16, 2026, Huntress reports. On November 12 one technician who joined the wrong session lost control of their workstation in about nine seconds. The ScreenConnect MSP guest attack exploits CVE-2026-84869 (CVSS 9.9, on CISA's exploited list since September 11), a flaw in the technician's own host client, so a patched server does not close it. That workstation usually reaches every customer the MSP supports, including yours.

Key takeaways

  • CVE-2026-84869 (CVSS 9.9, disclosed September 8, CISA KEV since September 11, 2026) lets a ScreenConnect guest make a technician's host client run a file without a prompt.
  • Huntress counted about 688,000 run-file requests on 42 MSP servers from 29 IPs between October 20 and November 16, 2026.
  • On November 12 a technician was compromised in about nine seconds and ended up with a rogue ScreenConnect client posing as Windows Update.
  • The fix is host client and Access agent version 26.6.5. A server update alone does not remediate; removing the TransferFiles permission is the interim control.
  • SCKit, seen November 11, is a separate campaign that planted rogue clients and session-cookie relays on hundreds of endpoints, apparently through a vulnerable server-side extension.
  • Customers should ask their MSP for per-technician client versions, a session database review and a hunt for unexplained remote-access clients.

What happened in the ScreenConnect MSP guest attack?

A flaw in ConnectWise ScreenConnect lets the person on the other end of a support session make the technician’s computer run a file. Huntress has watched attackers exploit it at scale against managed service providers. ConnectWise disclosed CVE-2026-84869 on September 8, 2026 and rated it 9.9 out of 10. CISA added it to its Known Exploited Vulnerabilities catalog three days later.

The mechanics are simple. A ScreenConnect Support or Access session has two sides: the host, which is the technician, and the guest, which is the machine being supported. The technician normally decides what moves and what runs. CVE-2026-84869 flips that. A guest can trigger file-transfer actions that the technician’s host client processes without authorization or a prompt, including running the file, potentially with elevated rights.

For a business that outsources IT, the target is what makes this uncomfortable. An MSP technician’s workstation is rarely just a laptop. It typically has the RMM console open, saved credentials in a browser or password manager, and a route into every client environment the MSP manages. An attacker who lands there does not need to beat your firewall. They can use your provider’s own tools, with the trust those tools carry.

The bug also lives in an awkward place. The vulnerable code runs on the technician’s machine, not on the ScreenConnect server. An MSP that updated its server in September and moved on may still have technicians connecting with old, exploitable clients. Huntress found exactly that.

How did attackers spray 42 MSP ScreenConnect servers?

The attacker needs two things: a guest session connected to the MSP’s ScreenConnect server, and a technician who joins it. The first part proved easier than it should have been. Huntress saw attacker-controlled guests sit on dozens of MSP servers for weeks. One kept reconnecting after a technician had queued it for uninstall.

A separate campaign Huntress reported on November 13 planted unauthorized guest sessions with crypto exchange names on more than 100 MSP ScreenConnect servers. Huntress did not see those sessions push files, but the names look like bait for a curious technician.

The second part is a numbers game. Technicians open sessions all day, and most are routine. The attacker’s guest asks the host client to run a file over and over, so whenever someone clicks in, the request lands within seconds. A vulnerable client shows no dialog to decline. No phishing email, no stolen password. Just patience and volume.

Huntress measured the volume. Between October 20 and November 16, 2026, attacker guests on 42 MSP ScreenConnect servers sent roughly 688,000 “run file” requests from 29 source IP addresses. The file names followed one pattern: update.bat, or update_ plus eight hexadecimal characters and .bat.

The busiest server took 70,414 requests over three weeks from 10 guest sessions using seven IPs. The spray guests there claimed client build 25.8.3.9466, the same build as the server, a cheap way to look unremarkable in the session list. They shared other habits an MSP can search for:

  • Blank session names, or names that look like default Windows hostnames.
  • No preview image, or a JPEG where a real screenshot would normally appear.
  • An empty guest public key, on about half the sessions on the busiest server.
  • The Company property (CustomProperty1) set to “crypto client” on 27 spray sessions across 17 servers. The crypto-exchange decoys from the other campaign left that field blank.

Huntress reads the eight-character suffix as a payload or wave identifier, not a per-victim tag. On November 11 the same sessions pushed one wave. On November 12 at 16:03 UTC, a new file name hit five MSP servers within two seconds of each other. Each received it between 585 and 5,671 times, and three were still receiving it on November 16.

That file was the exact first stage behind the compromise described below, and all five sessions that pushed it carried the “crypto client” tag. Huntress assesses with high confidence that the spray operator also ran the payload host and relay used in the break-in. It has not named a threat group.

How did a technician lose their workstation in nine seconds?

Forty-three minutes after the 16:03 wave began, a technician at one MSP joined a blank-named Access session. Process telemetry from the workstation puts the first file execution at 16:46:20.7 UTC and a rogue remote-access client running at 16:46:29.1. Roughly nine seconds, start to finish.

The chain went like this:

  1. The guest-pushed batch file ran from the host client’s Temp folder and relaunched itself through hidden PowerShell.
  2. It reported the new victim to the operator through a Telegram bot, with the bot token and chat ID Base64-encoded.
  3. It checked the Windows version, looked for an existing copy of its own client, and tested for admin rights.
  4. It used curl to download an MSI, with options tuned for flaky links: retries, resume, timeouts and a low-speed abort. Someone expected to fire this at many networks of uneven quality.
  5. It unpacked the MSI with msiexec in administrative mode, which extracts files to a folder instead of installing a product. No normal install record.
  6. It set the Secondary Logon service to start automatically, started it, and launched the payload with runas at the basic-user trust level, which runs a process with standard-user rights.
  7. It compiled a small piece of code on the fly with csc.exe.
  8. A rebranded ScreenConnect client, named to pass for Windows Update and placed in a per-wave folder under %LOCALAPPDATA%, connected to the attacker’s own relay.

That last step is the clever one. The attacker’s foothold is ScreenConnect itself. On an MSP workstation, ScreenConnect traffic is normal, so an analyst skimming network logs sees exactly the tool they expect.

Huntress published indicators, including the payload domain and source IP addresses. Take them from the Huntress write-up rather than secondhand copies, and treat them as a starting point: relay infrastructure changes.

What other attacks and near-misses did Huntress find?

November 12 was not the only time a guest got through. On November 13 at 01:49 UTC, a guest pushed a harmless text file to a second technician two seconds after they joined. About three seconds later their client ran it, and the file opened in Notepad++. Nothing else followed. The host went offline 18 minutes later.

A text file opening in an editor sounds trivial. For an attacker it is a canary: proof that delivery and execution work on a live target, without burning a real payload. That guest had been reconnecting to the MSP’s server since late October. The same IP pushed the marker file three times at a technician at a third MSP in late October. Huntress does not know whether that client ran it.

This guest behaved differently from the spray. It looked like a real, named Windows machine with a public key and a screenshot, which Huntress says fits a person driving it by hand. A minute earlier, the same technician had joined one of the crypto-named decoy sessions. Nothing came from it, but two campaigns had that technician in reach inside a minute.

The most targeted activity came in October on one MSP’s server. In three genuine customer support sessions, on October 23, 24 and 27, an attacker joined as an extra guest and asked the technician’s client to run a file through a relative path aimed at the technician’s Windows Startup folder. Anything there runs at the next sign-in, so the goal was persistence.

Each push hit a different technician, from two source IPs. Two were on vulnerable Windows clients. The third used the Mac client, where a Windows Startup path failed.

The near-misses show how thin the margin was. On October 19, a guest from the same /24 network as the canary pushed a batch file themed around lateral movement seven times at a technician on a vulnerable 24.2 client. The closest push arrived 4.3 seconds after the technician disconnected. On another server, a guest pushed a service payload 207 times, and the technician joined only after the last attempt.

Why is a ScreenConnect server update not enough?

Because the bug sits in the host client on each technician’s computer, and in Access agents, not in the server. ConnectWise lists 26.6.5 as the fixed version for both. Until every technician runs it, the server version on the admin dashboard tells you very little.

In the cases Huntress examined, many technicians were still on 22.x, 24.2 and 25.8 clients weeks after the fix shipped. We see the same drift in our own engagements: portable clients, a spare laptop that sat in a drawer, a contractor with their own install. Any one of them could be the way in.

ScreenConnect’s connection history records which client version each technician used for each session, so an MSP can pull that list in an afternoon and see who was exposed and when. For clients that cannot be updated at once, ConnectWise recommends removing the TransferFiles permission from roles and session groups. It breaks some workflows. It is still cheaper than a breach.

On the KEV listing: CISA’s deadlines bind US federal civilian agencies under Binding Operational Directive 22-01, not private companies. The listing still means CISA had evidence of exploitation by September 11. ScreenConnect has been here before: in February 2024 attackers mass-exploited CVE-2024-1709, an authentication bypass. Nothing ties the 2026 activity to those actors.

SCKit is a separate problem

On November 11, 2026, Huntress responded to worm-like activity it tracks as SCKit. The attacker used Run commands on cloud-hosted ScreenConnect panels to install rogue clients and session-cookie relays on hundreds of managed endpoints. Huntress says SCKit appears to abuse a vulnerable third-party server-side extension rather than CVE-2026-84869 itself.

Huntress calls the setup a “relay trap”: compromised endpoints staged to catch technicians who connect later. The guest file-run flaw attacks the technician from inside a session. SCKit seeds customer endpoints from the panel and waits for the technician to come to it.

Both can end the same way: a second remote-control channel the MSP does not own. A hunt that only checks client versions will miss SCKit, and one that only checks extensions will miss the guest pushes. Catching those beacons after the fact is a job for managed detection and response, not for a patch.

What should you ask your MSP about ScreenConnect this week?

Ask for evidence, not reassurance. “We’re patched” usually describes the server. Providers have had a difficult November, with N-able customers working through the N-central Able2Shell patch in the same weeks, so expect a busy team, but not a vague one. These questions get at the workstations and sessions that matter:

  • Client versions per technician. The host-client build each technician used, from ScreenConnect connection history, and the date each moved to 26.6.5. Include contractors and portable clients.
  • TransferFiles. If any technician is still below 26.6.5, was the TransferFiles permission removed, and when?
  • Suspicious sessions. Did anyone join a blank-named session, one with a default-looking hostname, or one tagged “crypto client”, especially on November 12 and 13? Are there crypto-exchange-named sessions nobody can explain?
  • Session database review. Have they searched for guest-initiated file runs that overlapped a connected technician?
  • Second clients. Have they scanned your endpoints and their technician PCs for ScreenConnect clients talking to a relay they do not own?
  • Extensions. Which third-party server extensions run on their instance, and have they been checked against the SCKit findings?
  • Notification. If a technician who supports you was compromised, when would you hear, and what would they share?

A competent MSP will answer most of this within a day or two and will not take offense. Until you have those answers, treat an unreviewed technician laptop as a possible path onto your own PCs.

What if something already ran through your MSP's remote tools?

Signs of a technician compromise

On the ScreenConnect server, the clearest sign is a guest making a technician’s client receive or run a file. Huntress points to the session database, App_Data\Session.db (work on a copy). Look for RanFiles (EventType 85) and SentFiles (EventType 86) events from guest connections (ProcessType 2) that overlap an open host connection (ProcessType 1).

A hit is not proof of compromise. It tells you which technician’s machine to examine, and at what timestamp. Pay extra attention to repeated update.bat requests in blank-named sessions and to any file path containing “../”.

On technician workstations, Huntress’s guidance comes down to:

  • The ScreenConnect host client spawning cmd.exe, powershell.exe, wscript.exe or notepad from its Temp directory.
  • msiexec in administrative mode with a target folder under %LOCALAPPDATA%, followed by runas.
  • curl downloading an MSI from a script the host client started.
  • Unexpected items in %APPDATA%\Microsoft\Windows\Start Menu\Programs\Startup.
  • The Secondary Logon service switched to automatic for no clear reason.
  • A ScreenConnect client with a Windows Update-style name in a user’s AppData folder.

On your own PCs and servers, the question is simpler, and the same check covers SCKit: is there any ScreenConnect client, or any remote-access tool, your MSP cannot account for? If nobody can answer with confidence, a threat hunting and compromise assessment gets you a straight answer faster than waiting for the attacker to show their hand.

How to respond

Treat a hit as an incident across every tenant that technician could reach, not as one infected laptop. Nine seconds is less time than it takes to read an alert, so assume the attacker held the technician’s access for as long as that rogue client stayed connected.

  1. Preserve first. Copy Session.db and pull endpoint telemetry before upgrades, cleanup scripts or uninstall jobs overwrite it. Huntress reconstructed the incidents above from exactly these records.
  2. Isolate the workstation, kill the rogue client, remove its AppData folder and MSI, and set Secondary Logon back to its normal start type.
  3. Map the reach. List every customer, RMM console, cloud tenant and admin account usable from that machine, then review actions taken since the first suspicious session.
  4. Rotate credentials reachable from the workstation, most privileged first: RMM, PSA, ScreenConnect, password vault, customer admin accounts.
  5. Hunt on customer endpoints for rogue ScreenConnect clients and other remote-access tools.
  6. Block Huntress’s published indicators as containment, not as an all-clear.

As the customer, ask your MSP for the timeline and the list of systems reviewed. If the answer is thin, or you already see odd logins or new admin accounts, bring in an independent data breach response team. An MSP investigating itself has an obvious conflict of interest, even when it acts in good faith.

What we check first

When we test MSP environments, the technician workstation is usually the shortest route to customer data. It is an ordinary Windows laptop, often with local admin rights, browsing the web, with a browser full of saved sessions. Few MSPs hold it to the controls they sell their clients.

Three checks tell us the most: the connection history filtered by client version, the session database query for guest file events overlapping a host connection, and a search for any ScreenConnect process whose relay is not the MSP’s own server. A clean IOC sweep does not close the case. Attackers rename relays and rebrand clients, which is exactly what happened here.

What is still unknown

Huntress has not named the group behind the spray or explained how the attacker guests reached 42 MSP servers. We do not know what the operator did after the November 12 compromise, how many customers sat behind that technician, or whether the hands-on canary operator is the same crew. The extension SCKit abused is also unnamed. Expect the counts to grow as more MSPs check their session databases.

If a technician who supports you may have been hit, or your MSP cannot show you clean answers, an incident response retainer puts an independent team on the case within hours.

← All articles

get in touch

Tell us what you're running

Scoping is free. We reply within one business day, and under 30 minutes for active incidents.