Operation ShadowCloud: MeshCentral Abuse Hides C2 in GitHub


What this means for your clients: Attackers are using a real remote-support tool to control computers and hiding their traffic inside GitHub, SharePoint and other cloud services, so ask your IT provider which remote tools are approved on your machines and who would notice a new one.
On November 17, Kaspersky's Global Research and Analysis Team (GReAT) published a two-part report and a press release on Operation ShadowCloud, an active cyber-espionage campaign. It abuses the legitimate MeshAgent utility to drop two previously undocumented backdoors, GitDoor and GraphDoor, which hide their command and control behind GitHub, Microsoft SharePoint and Ably.
Kaspersky says the campaign targets government organizations in Southeast Asia, Africa and Central Asia; the only specific victim it describes is a Taiwanese organization. In our judgment, most small businesses are not the targets, but the lesson travels: an RMM agent nobody approved is full remote control, and cloud C2 rarely appears on a list of bad domains.
How Operation ShadowCloud gets in and stays quiet
The entry point is unknown; Kaspersky says evidence points to the reuse of a prior infection. The MeshCentral Agent was hard-coded to a raw IP address, and once connected, the operators used it to drop three files side by side: a legitimate program, a malicious loader DLL and an encrypted payload, usually named config.ini.
That is DLL sideloading. Kaspersky found the Sandboxie start.exe in System32 beside a malicious sbiedll.dll, and a Python binary in a System32 subfolder beside python39.dll. The report also names jli.dll, WINMM.dll and libcef.dll as loaders and javaw.exe, pythonw.exe and zoomit.exe as hosts; the attackers started zoomit.exe as a service. The loader decrypts config.ini into memory and launches GitDoor or GraphDoor.
Two backdoors that live in your cloud allowlist
GitDoor uses a GitHub personal access token, pulling commands and plug-ins from one repository, uploading results as files to another, flagging sessions through issues, and posting to Ably. GraphDoor authenticates to Microsoft Graph as an application with an OAuth client-credentials grant, writes victim data to a SharePoint library and pulls commands from SharePoint lists.
Both download plug-ins on demand and inject shellcode in memory: a keylogger, screenshots, file transfer, command execution, service, firewall and registry changes, credential dumping from LSASS, and a search for WhatsApp decryption keys. Proxy plug-ins tunnel traffic to C2 servers through a modified Erfrp client or over Ably.
Kaspersky says it lacks sufficient information to definitively attribute this activity. It reports overlaps with Chinese-speaking groups FamousSparrow and CopperTurtle, whose implants shared staging folders with GitDoor, and says that does not prove a direct link.
Why MSPs should care
MeshCentral is the same class of tool many MSPs run daily. In our judgment, the agent nobody approved is the key signal, because everything after it hides in services clients allow. Kaspersky advises treating the introduction of a tool like MeshCentral as a suspicious event.
Lock down remote tools first
MeshCentral is not the flaw, so there is no patch. Our advice: inventory every RMM agent, allowlist only those the business installed, block other RMM installers with application control, and alert on any MeshAgent outside that list or talking to a server you do not run. Watch trusted binaries loading DLLs from odd folders beside a config.ini. Where your proxy allows, limit GitHub, SharePoint and Ably to the processes that need them.
If a host is hit, isolate it, preserve evidence, remove the agent, and from a clean system rotate every credential it could reach, including cloud and GitHub tokens.
How to hunt
We have not run these in a live SIEM; tune them to your field names and test before trusting a clean result. We start hunts at 2026-01-01T00:00:00Z, our choice, since Kaspersky found the campaign in early 2026; widen if retention allows. Splunk assumes Sysmon EID 1, 3, 7, 11, 13 and 22, Elastic assumes Elastic Defend, CrowdStrike assumes Falcon process, image-load, file-write, DNS and network events, and KQL assumes Defender for Endpoint. The campaign targets Windows; a MeshCentral server on Linux or macOS may have no telemetry. Every YOUR_ name is a placeholder.
Test 1: where MeshCentral and MeshAgent run
This 90-day inventory, our choice of window, lists hosts where MeshCentral or MeshAgent appears as a process, parent, file, registry key or connection, plus service or software inventory where kept. A long-running agent shows when it spawns a shell, writes files or reconnects, and inventory catches it idle. A renamed agent slips past name matching, so in our judgment also check file descriptions and signers. Approved RMM will fill the results.
Field mapping: Splunk adds the WinHostMon service snapshot if you collect it, Elastic includes already_running process events, Falcon checks processes only, with SyntheticProcessRollup2, and KQL adds the software inventory table.
Splunk
index=YOUR_SYSMON_INDEX (EventCode=1 OR EventCode=3 OR EventCode=11 OR EventCode=13) earliest=-90d latest=now
| eval subject=lower(coalesce(Image, "")." ".coalesce(CommandLine, "")." ".coalesce(TargetFilename, "")." ".coalesce(TargetObject, "")), parent=lower(coalesce(ParentImage, ""))
| eval where_seen=case(match(subject, "mesh ?(agent|central|service)"), "process, file, registry or network", match(parent, "mesh ?(agent|central|service)"), "child of a mesh process", true(), null())
| where isnotnull(where_seen)
| stats count AS events values(Image) AS images values(EventCode) AS event_codes max(_time) AS last_seen BY host where_seen
| append [search index=YOUR_WINDOWS_INDEX sourcetype=WinHostMon source=service earliest=-1d latest=now
| where match(lower(coalesce(Name, "")." ".coalesce(DisplayName, "")." ".coalesce(Path, "")), "mesh ?(agent|central|service)")
| stats latest(State) AS state values(Path) AS images max(_time) AS last_seen BY host
| eval where_seen="service inventory"]
| convert ctime(last_seen)Elastic
FROM logs-endpoint.events.*
| WHERE @timestamp >= NOW() - 90 days
| EVAL subject = TO_LOWER(CONCAT(COALESCE(process.executable, ""), " ", COALESCE(process.command_line, ""), " ", COALESCE(file.path, ""), " ", COALESCE(registry.path, ""))), parent = TO_LOWER(COALESCE(process.parent.executable, ""))
| EVAL where_seen = CASE(subject RLIKE ".*mesh ?(agent|central|service).*", "process, file, registry or network", parent RLIKE ".*mesh ?(agent|central|service).*", "child of a mesh process", NULL)
| WHERE where_seen IS NOT NULL
| STATS events = COUNT(*), images = VALUES(process.executable), actions = VALUES(event.action), last_seen = MAX(@timestamp) BY host.name, where_seen
| SORT host.nameCrowdStrike
// set the time picker to the last 90 days
#event_simpleName=/^(ProcessRollup2|SyntheticProcessRollup2)$/
| case {
ImageFileName=/mesh ?(agent|central|service)/i | where_seen := "process" ;
CommandLine=/mesh ?(agent|central|service)/i | where_seen := "process" ;
ParentBaseFileName=/mesh ?(agent|central|service)/i | where_seen := "child of a mesh process" }
| groupBy([aid, where_seen], function=[count(as=events), collect([ImageFileName, #event_simpleName]), max(@timestamp, as=last_seen)])Microsoft KQL
let mesh = @"(?i)mesh ?(agent|central|service)";
union
(DeviceProcessEvents
| where Timestamp > ago(90d)
| extend WhereSeen = case(strcat(FolderPath, " ", ProcessCommandLine) matches regex mesh, "process", strcat(InitiatingProcessFolderPath, " ", InitiatingProcessCommandLine) matches regex mesh, "child of a mesh process", ""), Item = FolderPath),
(DeviceNetworkEvents
| where Timestamp > ago(90d)
| extend WhereSeen = iff(InitiatingProcessFolderPath matches regex mesh, "network", ""), Item = InitiatingProcessFolderPath),
(DeviceFileEvents
| where Timestamp > ago(90d)
| extend WhereSeen = iff(strcat(FolderPath, " ", InitiatingProcessFolderPath) matches regex mesh, "file", ""), Item = FolderPath),
(DeviceRegistryEvents
| where Timestamp > ago(90d)
| extend WhereSeen = iff(RegistryKey matches regex mesh, "registry", ""), Item = RegistryKey),
(DeviceTvmSoftwareInventory
| extend Timestamp = now(), WhereSeen = iff(SoftwareName matches regex @"(?i)mesh.?(agent|central)", "software inventory", ""), Item = SoftwareName)
| where isnotempty(WhereSeen)
| summarize Events = count(), Items = make_set(Item, 20), LastSeen = max(Timestamp) by DeviceName, WhereSeen
| order by DeviceName ascHunt 1: MeshAgent where it should not be
This flags MeshAgent on a host outside your approved RMM list or connecting outside your server range, plus the MeshAgent C2 address and MD5 Kaspersky published.
Field mapping: Falcon network events carry no process name, so CrowdStrike skips the server-range branch; Splunk needs Sysmon MD5 hashing.
False positives: your own MeshCentral hits until you fill in the approved lists.
Splunk
index=YOUR_SYSMON_INDEX (EventCode=1 OR EventCode=3) earliest=1767225600 latest=now
| eval img=lower(Image), md5=lower(replace(coalesce(Hashes, ""), ".*MD5=([0-9A-Fa-f]{32}).*", "\1"))
| eval is_mesh=if(match(img, "mesh ?(agent|central|service)"), 1, 0)
| eval reason=case(DestinationIp=="27.102.66.162", "published MeshAgent C2", md5=="74b025084026d8d9a64661bf995d835b", "published MeshAgent MD5", is_mesh==1 AND EventCode==3 AND NOT cidrmatch("YOUR_MESH_SERVER_CIDR", DestinationIp), "mesh agent to unapproved server", is_mesh==1 AND NOT match(host, "YOUR_APPROVED_RMM_HOST_REGEX"), "mesh agent on unapproved host", true(), null())
| where isnotnull(reason)
| stats count AS events values(Image) AS images values(DestinationIp) AS dests min(_time) AS first_seen max(_time) AS last_seen BY host reason
| convert ctime(first_seen) ctime(last_seen)Elastic
FROM logs-endpoint.events.process-*, logs-endpoint.events.network-*
| WHERE @timestamp >= TO_DATETIME("2026-01-01T00:00:00.000Z")
| EVAL img = TO_LOWER(process.executable), md5 = TO_LOWER(process.hash.md5), dst = TO_STRING(destination.ip)
| EVAL is_mesh = img RLIKE ".*mesh ?(agent|central|service).*"
| EVAL reason = CASE(dst == "27.102.66.162", "published MeshAgent C2",
md5 == "74b025084026d8d9a64661bf995d835b", "published MeshAgent MD5",
is_mesh AND event.dataset == "endpoint.events.network" AND NOT CIDR_MATCH(destination.ip, "YOUR_MESH_SERVER_CIDR"), "mesh agent to unapproved server",
is_mesh AND NOT host.name IN ("YOUR_APPROVED_RMM_HOST_1", "YOUR_APPROVED_RMM_HOST_2"), "mesh agent on unapproved host", NULL)
| WHERE reason IS NOT NULL
| STATS events = COUNT(*), images = VALUES(process.executable), dests = VALUES(dst), first_seen = MIN(@timestamp), last_seen = MAX(@timestamp) BY host.name, reasonCrowdStrike
// set the time picker to start on or before 2026-01-01
// REPLACE_WITH_APPROVED_AID is a placeholder: replace both values with your approved RMM host aids before running
#event_simpleName=/^(ProcessRollup2|SyntheticProcessRollup2|NetworkConnectIP4)$/
| test(@timestamp >= 1767225600000)
| case {
#event_simpleName=NetworkConnectIP4 | RemoteAddressIP4="27.102.66.162" | reason := "published MeshAgent C2" ;
#event_simpleName=/ProcessRollup2$/ | MD5HashData=/^74b025084026d8d9a64661bf995d835b$/i | reason := "published MeshAgent MD5" ;
#event_simpleName=/ProcessRollup2$/ | ImageFileName=/mesh ?(agent|central|service)/i | !in(field=aid, values=["REPLACE_WITH_APPROVED_AID_1", "REPLACE_WITH_APPROVED_AID_2"]) | reason := "mesh agent on unapproved host" }
| default(field=[ImageFileName, RemoteAddressIP4], value="n/a")
| groupBy([aid, reason], function=[count(as=events), collect([ImageFileName, RemoteAddressIP4]), min(@timestamp, as=first_seen), max(@timestamp, as=last_seen)])Microsoft KQL
let mesh = @"(?i)mesh ?(agent|central|service)";
union
(DeviceNetworkEvents
| where Timestamp >= datetime(2026-01-01T00:00:00Z)
| extend IsMesh = InitiatingProcessFolderPath matches regex mesh, Image = InitiatingProcessFolderPath, RIP = replace_string(RemoteIP, "::ffff:", "")
| extend Reason = case(RIP == "27.102.66.162", "published MeshAgent C2", IsMesh and not(ipv4_is_in_range(RIP, "YOUR_MESH_SERVER_CIDR")), "mesh agent to unapproved server", "")),
(DeviceProcessEvents
| where Timestamp >= datetime(2026-01-01T00:00:00Z)
| extend Image = FolderPath
| extend Reason = case(MD5 =~ "74b025084026d8d9a64661bf995d835b", "published MeshAgent MD5", FolderPath matches regex mesh and DeviceName !in~ ("YOUR_APPROVED_RMM_HOST_1", "YOUR_APPROVED_RMM_HOST_2"), "mesh agent on unapproved host", ""))
| where isnotempty(Reason)
| summarize Events = count(), Images = make_set(Image, 20), Dests = make_set(RemoteIP, 20), FirstSeen = min(Timestamp), LastSeen = max(Timestamp) by DeviceName, ReasonHunt 2: sideloaded loader DLLs
This finds javaw.exe, start.exe, pythonw.exe, zoomit.exe or dtsapo4service.exe loading jli.dll, sbiedll.dll, python39.dll, WINMM.dll or libcef.dll from its own folder, plus same-folder loads of them from ProgramData, a user profile, Windows\Temp, System32\dts or PDF24. Excluding other Program Files folders is our choice.
Field mapping: Defender and Falcon record image loads selectively, and CrowdStrike checks only the DLL folder.
False positives: per-user Python and portable Java will appear and need baselining; check the signer.
Splunk
index=YOUR_SYSMON_INDEX EventCode=7 earliest=1767225600 latest=now
| eval img=lower(Image), dll=lower(ImageLoaded)
| eval proc=replace(img, "^.*\\\\", ""), pdir=replace(img, "[^\\\\]+$", ""), dname=replace(dll, "^.*\\\\", ""), ddir=replace(dll, "[^\\\\]+$", "")
| where in(dname, "jli.dll", "sbiedll.dll", "python39.dll", "winmm.dll", "libcef.dll") AND pdir==ddir
| where in(proc, "javaw.exe", "start.exe", "pythonw.exe", "zoomit.exe", "dtsapo4service.exe") OR match(ddir, "\\\\(programdata|users|windows\\\\temp|windows\\\\system32\\\\dts)\\\\|\\\\program files\\\\pdf24\\\\")
| where NOT (match(ddir, "^c:\\\\program files( [(]x86[)])?\\\\") AND NOT match(ddir, "\\\\pdf24\\\\"))
| stats count AS loads values(Image) AS hosts values(ImageLoaded) AS dlls values(Signed) AS signed min(_time) AS first_seen max(_time) AS last_seen BY host
| convert ctime(first_seen) ctime(last_seen)Elastic
FROM logs-endpoint.events.library-*
| WHERE @timestamp >= TO_DATETIME("2026-01-01T00:00:00.000Z")
| EVAL img = TO_LOWER(process.executable), dll = TO_LOWER(dll.path), proc = TO_LOWER(process.name), dname = TO_LOWER(dll.name)
| WHERE dname IN ("jli.dll", "sbiedll.dll", "python39.dll", "winmm.dll", "libcef.dll")
| EVAL pdir = SUBSTRING(img, 1, LENGTH(img) - LENGTH(proc)), ddir = SUBSTRING(dll, 1, LENGTH(dll) - LENGTH(dname))
| WHERE pdir == ddir
| WHERE proc IN ("javaw.exe", "start.exe", "pythonw.exe", "zoomit.exe", "dtsapo4service.exe")
OR ddir RLIKE """.*\\(programdata|users|windows\\temp|windows\\system32\\dts)\\.*"""
OR ddir RLIKE """.*\\program files\\pdf24\\.*"""
| WHERE NOT (ddir RLIKE """c:\\program files( \(x86\))?\\.*""" AND NOT ddir RLIKE """.*\\pdf24\\.*""")
| STATS loads = COUNT(*), hosts = VALUES(process.executable), dlls = VALUES(dll.path), trusted = VALUES(dll.code_signature.trusted), first_seen = MIN(@timestamp), last_seen = MAX(@timestamp) BY host.nameCrowdStrike
// set the time picker to start on or before 2026-01-01
#event_simpleName=/^(ImageHash|ClassifiedModuleLoad)$/
| test(@timestamp >= 1767225600000)
| ImageFileName=/\\(jli|sbiedll|python39|winmm|libcef)\.dll$/i
| case {
ImageFileName=/\\Program Files\\PDF24\\/i | keep := 1 ;
ImageFileName=/\\Program Files( \(x86\))?\\/i | keep := 0 ;
* | keep := 1 }
| keep=1
| ImageFileName=/\\(ProgramData|Users|Windows\\Temp|Windows\\System32\\dts|Program Files\\PDF24)\\|\\Windows\\System32\\(jli|sbiedll|python39|libcef)\.dll$/i
| groupBy([aid, ImageFileName], function=[count(as=loads), collect([MD5HashData, ContextProcessId]), min(@timestamp, as=first_seen), max(@timestamp, as=last_seen)])Microsoft KQL
DeviceImageLoadEvents
| where Timestamp >= datetime(2026-01-01T00:00:00Z)
| where FileName in~ ("jli.dll", "sbiedll.dll", "python39.dll", "winmm.dll", "libcef.dll")
| extend Proc = tolower(InitiatingProcessFileName), PPath = tolower(InitiatingProcessFolderPath), DPath = tolower(FolderPath)
| extend PDir = iff(PPath endswith Proc, extract(@"^(.*\\)", 1, PPath), strcat(PPath, @"\")),
DDir = iff(DPath endswith tolower(FileName), extract(@"^(.*\\)", 1, DPath), strcat(DPath, @"\"))
| where PDir == DDir
| where Proc in ("javaw.exe", "start.exe", "pythonw.exe", "zoomit.exe", "dtsapo4service.exe") or DDir matches regex @"\\(programdata|users|windows\\temp|windows\\system32\\dts)\\|\\program files\\pdf24\\"
| where not(DDir matches regex @"^c:\\program files( \(x86\))?\\" and not(DDir contains @"\pdf24\"))
| summarize Loads = count(), Hosts = make_set(InitiatingProcessFolderPath, 20), Dlls = make_set(FolderPath, 20), FirstSeen = min(Timestamp), LastSeen = max(Timestamp) by DeviceNameHunt 3: config.ini staging and zoomit as a service
This finds config.ini written in System32, System32\dts or the GitDoor staging folders ProgramData\excint and PDF24, and zoomit.exe started by services.exe.
Field mapping: not every tool logs .ini writes, Falcon especially, so silence is weak.
False positives: some vendors drop config.ini; we expect a zoomit service to be rare.
Splunk
index=YOUR_SYSMON_INDEX (EventCode=1 OR EventCode=11) earliest=1767225600 latest=now
| eval tf=lower(TargetFilename), proc=lower(replace(Image, "^.*\\\\", "")), parent=lower(replace(ParentImage, "^.*\\\\", ""))
| eval reason=case(EventCode==11 AND match(tf, "^c:\\\\windows\\\\system32\\\\(dts\\\\(.*\\\\)?)?config[.]ini$|\\\\programdata\\\\excint\\\\(.*\\\\)?config[.]ini$|\\\\program files\\\\pdf24\\\\config[.]ini$"), "config.ini in an observed staging folder", EventCode==1 AND match(proc, "^zoomit(64)?[.]exe$") AND parent=="services.exe", "zoomit started as a service", true(), null())
| where isnotnull(reason)
| stats count AS events values(TargetFilename) AS files values(CommandLine) AS commands values(Image) AS images min(_time) AS first_seen BY host reason
| convert ctime(first_seen)Elastic
FROM logs-endpoint.events.process-*, logs-endpoint.events.file-*
| WHERE @timestamp >= TO_DATETIME("2026-01-01T00:00:00.000Z")
| EVAL fp = TO_LOWER(file.path), proc = TO_LOWER(process.name), parent = TO_LOWER(process.parent.name)
| EVAL reason = CASE(event.dataset == "endpoint.events.file" AND fp RLIKE """(c:\\windows\\system32\\(dts\\(.*\\)?)?|.*\\programdata\\excint\\(.*\\)?|.*\\program files\\pdf24\\)config\.ini""", "config.ini in an observed staging folder",
event.dataset == "endpoint.events.process" AND event.action == "start" AND proc RLIKE """zoomit(64)?\.exe""" AND parent == "services.exe", "zoomit started as a service", NULL)
| WHERE reason IS NOT NULL
| STATS events = COUNT(*), files = VALUES(file.path), commands = VALUES(process.command_line), images = VALUES(process.executable), first_seen = MIN(@timestamp) BY host.name, reasonCrowdStrike
// set the time picker to start on or before 2026-01-01
#event_simpleName=/^(ProcessRollup2|.*Written)$/
| test(@timestamp >= 1767225600000)
| case {
#event_simpleName=/Written$/ | TargetFileName=/(\\Windows\\System32\\(dts\\(.*\\)?)?|\\ProgramData\\excint\\(.*\\)?|\\Program Files\\PDF24\\)config\.ini$/i | reason := "config.ini in an observed staging folder" ;
#event_simpleName=ProcessRollup2 | FileName=/^zoomit(64)?\.exe$/i | ParentBaseFileName=/^services\.exe$/i | reason := "zoomit started as a service" }
| default(field=[TargetFileName, CommandLine], value="n/a")
| groupBy([aid, reason], function=[count(as=events), collect([TargetFileName, CommandLine]), min(@timestamp, as=first_seen)])Microsoft KQL
union
(DeviceFileEvents
| where Timestamp >= datetime(2026-01-01T00:00:00Z)
| where ActionType in ("FileCreated", "FileRenamed", "FileModified") and FileName =~ "config.ini"
| where FolderPath matches regex @"(?i)^c:\\windows\\system32\\(dts\\(.*\\)?)?config\.ini$|\\programdata\\excint\\(.*\\)?config\.ini$|\\program files\\pdf24\\config\.ini$"
| extend Reason = "config.ini in an observed staging folder", Detail = FolderPath),
(DeviceProcessEvents
| where Timestamp >= datetime(2026-01-01T00:00:00Z)
| where FileName matches regex @"(?i)^zoomit(64)?\.exe$" and InitiatingProcessFileName =~ "services.exe"
| extend Reason = "zoomit started as a service", Detail = strcat(FolderPath, " ", ProcessCommandLine))
| summarize Events = count(), Details = make_set(Detail, 20), FirstSeen = min(Timestamp) by DeviceName, ReasonHunt 4: non-browser processes calling GitHub, Graph, SharePoint or Ably
This finds processes other than browsers, Office, OneDrive, Teams, git and common IDEs looking up api.github.com, login.microsoftonline.com, graph.microsoft.com, SharePoint, Ably or upload.gofile.io. The exclusion list is ours.
Field mapping: Falcon only names processes started in the search range, which we open a day early; Hunt 6 catches the published domains without that join. Behind TLS inspection, client-credentials token requests from a non-browser process are a stronger signal.
False positives: sync clients, updaters and scripts will appear; baseline first.
Splunk
index=YOUR_SYSMON_INDEX EventCode=22 earliest=1767225600 latest=now
| eval q=lower(QueryName), proc=lower(replace(Image, "^.*\\\\", ""))
| where match(q, "^(api[.]github[.]com|login[.]microsoftonline[.]com|graph[.]microsoft[.]com|[a-z0-9-]+[.]sharepoint[.]com|([a-z0-9-]+[.])*ably[.](io|net)|upload[.]gofile[.]io)$")
| where NOT match(proc, "(?i)^(chrome|msedge|msedgewebview2|firefox|brave|iexplore|outlook|winword|excel|powerpnt|onenote|ms-teams|teams|onedrive|git|git-remote-https|code|devenv|githubdesktop|gh|YOUR_APPROVED_AUTOMATION)[.]exe$")
| stats count AS lookups values(q) AS domains values(Image) AS images min(_time) AS first_seen max(_time) AS last_seen BY host proc
| convert ctime(first_seen) ctime(last_seen)Elastic
// enter YOUR_APPROVED_AUTOMATION.exe in lowercase
FROM logs-endpoint.events.network-*
| WHERE @timestamp >= TO_DATETIME("2026-01-01T00:00:00.000Z") AND dns.question.name IS NOT NULL
| EVAL q = TO_LOWER(dns.question.name), proc = TO_LOWER(process.name)
| WHERE q RLIKE "(api[.]github[.]com|login[.]microsoftonline[.]com|graph[.]microsoft[.]com|[a-z0-9-]+[.]sharepoint[.]com|([a-z0-9-]+[.])*ably[.](io|net)|upload[.]gofile[.]io)"
| WHERE NOT proc IN ("chrome.exe", "msedge.exe", "msedgewebview2.exe", "firefox.exe", "brave.exe", "iexplore.exe", "outlook.exe", "winword.exe", "excel.exe", "powerpnt.exe", "onenote.exe", "ms-teams.exe", "teams.exe", "onedrive.exe", "git.exe", "git-remote-https.exe", "code.exe", "devenv.exe", "githubdesktop.exe", "gh.exe", "YOUR_APPROVED_AUTOMATION.exe")
| STATS lookups = COUNT(*), domains = VALUES(q), images = VALUES(process.executable), first_seen = MIN(@timestamp), last_seen = MAX(@timestamp) BY host.name, procCrowdStrike
// set the time picker to start on or before 2025-12-31, one day before the hunt start
#event_simpleName=/^(ProcessRollup2|DnsRequest)$/
| test(@timestamp >= 1767139200000)
| case {
#event_simpleName=ProcessRollup2 | FileName!=/^(chrome|msedge|msedgewebview2|firefox|brave|iexplore|outlook|winword|excel|powerpnt|onenote|ms-teams|teams|onedrive|git|git-remote-https|code|devenv|githubdesktop|gh|YOUR_APPROVED_AUTOMATION)\.exe$/i | ContextProcessId := TargetProcessId | proc := FileName ;
#event_simpleName=DnsRequest | test(@timestamp >= 1767225600000) | DomainName=/^(api\.github\.com|login\.microsoftonline\.com|graph\.microsoft\.com|[a-z0-9-]+\.sharepoint\.com|([a-z0-9-]+\.)*ably\.(io|net)|upload\.gofile\.io)$/i | dom := lower(DomainName) }
| selfJoinFilter(field=[aid, ContextProcessId], where=[{#event_simpleName=ProcessRollup2}, {#event_simpleName=DnsRequest}])
| groupBy([aid, ContextProcessId], function=[count(field=proc, as=procs), count(field=dom, as=lookups), collect([proc, dom]), min(@timestamp, as=first_seen)], limit=max)
| test(procs > 0)
| test(lookups > 0)Microsoft KQL
let allow = dynamic(["chrome.exe", "msedge.exe", "msedgewebview2.exe", "firefox.exe", "brave.exe", "iexplore.exe", "outlook.exe", "winword.exe", "excel.exe", "powerpnt.exe", "onenote.exe", "ms-teams.exe", "teams.exe", "onedrive.exe", "git.exe", "git-remote-https.exe", "code.exe", "devenv.exe", "githubdesktop.exe", "gh.exe", "YOUR_APPROVED_AUTOMATION.exe"]);
DeviceNetworkEvents
| where Timestamp >= datetime(2026-01-01T00:00:00Z)
| where isnotempty(RemoteUrl)
| extend Host = tolower(tostring(split(replace_regex(RemoteUrl, @"^[a-zA-Z]+://", ""), "/")[0]))
| extend Host = tostring(split(Host, ":")[0])
| where Host matches regex @"^(api\.github\.com|login\.microsoftonline\.com|graph\.microsoft\.com|[a-z0-9-]+\.sharepoint\.com|([a-z0-9-]+\.)*ably\.(io|net)|upload\.gofile\.io)$"
| where InitiatingProcessFileName !in~ (allow)
| summarize Connections = count(), Hosts = make_set(Host, 20), Paths = make_set(InitiatingProcessFolderPath, 10), FirstSeen = min(Timestamp), LastSeen = max(Timestamp) by DeviceName, InitiatingProcessFileNameHunt 5: published hashes
This matches the 11 campaign MD5 values Kaspersky published and four from its attribution section: a FamousSparrow DLL, a CopperTurtle executable and two GitDoor DLLs sharing their folders. The four legitimate sideload hosts and the renamed Avast file stay in the table but out of the query, our choice, since a match only proves a known-good file.
Field mapping: Splunk searches EDR, antivirus and Sysmon data by raw term, and in-memory plug-ins may never be hashed.
False positives: we expect none; verify raw-term hits.
Splunk
index=YOUR_EDR_AV_AND_SYSMON_INDEXES earliest=1767225600 latest=now (74b025084026d8d9a64661bf995d835b OR 32151547d912488ac5242731057170a3 OR e9d4081fac4a5111e6856f333d490439 OR e63ec733393e4e7e3424cbff918003a3 OR 3896d44bc6d517291c60ee3f70543a31 OR b930fb3ab14fe45b378d5acfa6e0a763 OR 6a54f8008ee659b6923abf5dba2b2cc1 OR 552f0a9cc5fc3755721fa49be05e8035 OR 14ec33ea808e697870d00c5e434c2952 OR d89b2fe0759df2d534e7a728700a6f8c OR b4956b230d296cb4b015b42b11d80e32 OR 733f8002aa9531b25384c0014d15cc1f OR 91bee19d31d32346cd4d1febb4308606 OR 45b3aa39bc33528650d9181cd946b2df OR 9a5e0ab7466594dcf46943c413db7b81)
| rex max_match=0 "(?i)(?<md5_hit>74b025084026d8d9a64661bf995d835b|32151547d912488ac5242731057170a3|e9d4081fac4a5111e6856f333d490439|e63ec733393e4e7e3424cbff918003a3|3896d44bc6d517291c60ee3f70543a31|b930fb3ab14fe45b378d5acfa6e0a763|6a54f8008ee659b6923abf5dba2b2cc1|552f0a9cc5fc3755721fa49be05e8035|14ec33ea808e697870d00c5e434c2952|d89b2fe0759df2d534e7a728700a6f8c|b4956b230d296cb4b015b42b11d80e32|733f8002aa9531b25384c0014d15cc1f|91bee19d31d32346cd4d1febb4308606|45b3aa39bc33528650d9181cd946b2df|9a5e0ab7466594dcf46943c413db7b81)"
| eval md5_hit=lower(md5_hit)
| stats count AS events values(md5_hit) AS md5s values(sourcetype) AS sources min(_time) AS first_seen max(_time) AS last_seen BY host
| convert ctime(first_seen) ctime(last_seen)Elastic
FROM logs-endpoint.*
| WHERE @timestamp >= TO_DATETIME("2026-01-01T00:00:00.000Z")
| EVAL file_md5 = TO_LOWER(file.hash.md5), process_md5 = TO_LOWER(process.hash.md5), dll_md5 = TO_LOWER(dll.hash.md5)
| WHERE file_md5 IN ("74b025084026d8d9a64661bf995d835b", "32151547d912488ac5242731057170a3", "e9d4081fac4a5111e6856f333d490439", "e63ec733393e4e7e3424cbff918003a3", "3896d44bc6d517291c60ee3f70543a31", "b930fb3ab14fe45b378d5acfa6e0a763", "6a54f8008ee659b6923abf5dba2b2cc1", "552f0a9cc5fc3755721fa49be05e8035", "14ec33ea808e697870d00c5e434c2952", "d89b2fe0759df2d534e7a728700a6f8c", "b4956b230d296cb4b015b42b11d80e32", "733f8002aa9531b25384c0014d15cc1f", "91bee19d31d32346cd4d1febb4308606", "45b3aa39bc33528650d9181cd946b2df", "9a5e0ab7466594dcf46943c413db7b81")
OR process_md5 IN ("74b025084026d8d9a64661bf995d835b", "32151547d912488ac5242731057170a3", "e9d4081fac4a5111e6856f333d490439", "e63ec733393e4e7e3424cbff918003a3", "3896d44bc6d517291c60ee3f70543a31", "b930fb3ab14fe45b378d5acfa6e0a763", "6a54f8008ee659b6923abf5dba2b2cc1", "552f0a9cc5fc3755721fa49be05e8035", "14ec33ea808e697870d00c5e434c2952", "d89b2fe0759df2d534e7a728700a6f8c", "b4956b230d296cb4b015b42b11d80e32", "733f8002aa9531b25384c0014d15cc1f", "91bee19d31d32346cd4d1febb4308606", "45b3aa39bc33528650d9181cd946b2df", "9a5e0ab7466594dcf46943c413db7b81")
OR dll_md5 IN ("74b025084026d8d9a64661bf995d835b", "32151547d912488ac5242731057170a3", "e9d4081fac4a5111e6856f333d490439", "e63ec733393e4e7e3424cbff918003a3", "3896d44bc6d517291c60ee3f70543a31", "b930fb3ab14fe45b378d5acfa6e0a763", "6a54f8008ee659b6923abf5dba2b2cc1", "552f0a9cc5fc3755721fa49be05e8035", "14ec33ea808e697870d00c5e434c2952", "d89b2fe0759df2d534e7a728700a6f8c", "b4956b230d296cb4b015b42b11d80e32", "733f8002aa9531b25384c0014d15cc1f", "91bee19d31d32346cd4d1febb4308606", "45b3aa39bc33528650d9181cd946b2df", "9a5e0ab7466594dcf46943c413db7b81")
| STATS events = COUNT(*), files = VALUES(file.path), images = VALUES(process.executable), dlls = VALUES(dll.path), first_seen = MIN(@timestamp) BY host.name, file_md5, process_md5, dll_md5CrowdStrike
// set the time picker to start on or before 2026-01-01
MD5HashData=*
| test(@timestamp >= 1767225600000)
| in(field=MD5HashData, values=["74b025084026d8d9a64661bf995d835b", "32151547d912488ac5242731057170a3", "e9d4081fac4a5111e6856f333d490439", "e63ec733393e4e7e3424cbff918003a3", "3896d44bc6d517291c60ee3f70543a31", "b930fb3ab14fe45b378d5acfa6e0a763", "6a54f8008ee659b6923abf5dba2b2cc1", "552f0a9cc5fc3755721fa49be05e8035", "14ec33ea808e697870d00c5e434c2952", "d89b2fe0759df2d534e7a728700a6f8c", "b4956b230d296cb4b015b42b11d80e32", "733f8002aa9531b25384c0014d15cc1f", "91bee19d31d32346cd4d1febb4308606", "45b3aa39bc33528650d9181cd946b2df", "9a5e0ab7466594dcf46943c413db7b81"], ignoreCase=true)
| default(field=[ImageFileName, TargetFileName], value="n/a")
| groupBy([aid, MD5HashData, #event_simpleName], function=[count(as=events), collect([ImageFileName, TargetFileName]), min(@timestamp, as=first_seen)], limit=max)Microsoft KQL
let md5s = dynamic(["74b025084026d8d9a64661bf995d835b", "32151547d912488ac5242731057170a3", "e9d4081fac4a5111e6856f333d490439", "e63ec733393e4e7e3424cbff918003a3", "3896d44bc6d517291c60ee3f70543a31", "b930fb3ab14fe45b378d5acfa6e0a763", "6a54f8008ee659b6923abf5dba2b2cc1", "552f0a9cc5fc3755721fa49be05e8035", "14ec33ea808e697870d00c5e434c2952", "d89b2fe0759df2d534e7a728700a6f8c", "b4956b230d296cb4b015b42b11d80e32", "733f8002aa9531b25384c0014d15cc1f", "91bee19d31d32346cd4d1febb4308606", "45b3aa39bc33528650d9181cd946b2df", "9a5e0ab7466594dcf46943c413db7b81"]);
union withsource = Table
(DeviceProcessEvents
| where Timestamp >= datetime(2026-01-01T00:00:00Z)
| where MD5 in~ (md5s) or InitiatingProcessMD5 in~ (md5s)
| extend Hit = iff(MD5 in~ (md5s), MD5, InitiatingProcessMD5), Path = iff(MD5 in~ (md5s), FolderPath, InitiatingProcessFolderPath)),
(DeviceFileEvents
| where Timestamp >= datetime(2026-01-01T00:00:00Z)
| where MD5 in~ (md5s)
| extend Hit = MD5, Path = FolderPath),
(DeviceImageLoadEvents
| where Timestamp >= datetime(2026-01-01T00:00:00Z)
| where MD5 in~ (md5s)
| extend Hit = MD5, Path = FolderPath)
| summarize Events = count(), Paths = make_set(Path, 20), FirstSeen = min(Timestamp) by DeviceName, Hit = tolower(Hit), TableHunt 6: published C2 addresses and domains
This matches the MeshAgent C2 and 22 proxy C2 addresses Kaspersky published, plus lookups for its two SharePoint sites, upload.gofile.io and two domains it says also served Cobalt Strike.
Field mapping: address hits need connection logs, and domain hits need DNS logging on the host.
False positives: upload.gofile.io is a legitimate service; check the process.
Splunk
index=YOUR_SYSMON_INDEX (EventCode=3 OR EventCode=22) earliest=1767225600 latest=now
| eval q=lower(QueryName)
| eval indicator=case(EventCode==3 AND in(DestinationIp, "27.102.66.162", "104.194.154.28", "216.73.158.72", "122.147.252.16", "103.106.230.33", "123.58.205.105", "223.165.4.203", "172.86.66.19", "103.214.68.20", "216.146.25.173", "103.137.63.39", "103.227.224.189", "104.194.152.173", "66.42.52.137", "153.75.94.148", "45.59.124.164", "104.194.153.161", "153.75.95.27", "216.73.158.117", "154.205.154.209", "43.134.62.111", "38.54.17.113", "38.52.17.113"), DestinationIp, EventCode==22 AND match(q, "(^|[.])(buildersemp[.]sharepoint[.]com|cskautogo[.]sharepoint[.]com|upload[.]gofile[.]io|twnic[.]top|ekmt[.]tw)$"), q, true(), null())
| where isnotnull(indicator)
| stats count AS events values(Image) AS images min(_time) AS first_seen max(_time) AS last_seen BY host indicator
| convert ctime(first_seen) ctime(last_seen)Elastic
FROM logs-endpoint.events.network-*
| WHERE @timestamp >= TO_DATETIME("2026-01-01T00:00:00.000Z")
| EVAL dst = TO_STRING(destination.ip), q = TO_LOWER(dns.question.name)
| EVAL indicator = CASE(dst IN ("27.102.66.162", "104.194.154.28", "216.73.158.72", "122.147.252.16", "103.106.230.33", "123.58.205.105", "223.165.4.203", "172.86.66.19", "103.214.68.20", "216.146.25.173", "103.137.63.39", "103.227.224.189", "104.194.152.173", "66.42.52.137", "153.75.94.148", "45.59.124.164", "104.194.153.161", "153.75.95.27", "216.73.158.117", "154.205.154.209", "43.134.62.111", "38.54.17.113", "38.52.17.113"), dst, q RLIKE "(.*[.])?(buildersemp[.]sharepoint[.]com|cskautogo[.]sharepoint[.]com|upload[.]gofile[.]io|twnic[.]top|ekmt[.]tw)", q, NULL)
| WHERE indicator IS NOT NULL
| STATS events = COUNT(*), images = VALUES(process.executable), first_seen = MIN(@timestamp), last_seen = MAX(@timestamp) BY host.name, indicatorCrowdStrike
// set the time picker to start on or before 2026-01-01
#event_simpleName=/^(NetworkConnectIP4|DnsRequest)$/
| test(@timestamp >= 1767225600000)
| case {
#event_simpleName=NetworkConnectIP4 | in(field=RemoteAddressIP4, values=["27.102.66.162", "104.194.154.28", "216.73.158.72", "122.147.252.16", "103.106.230.33", "123.58.205.105", "223.165.4.203", "172.86.66.19", "103.214.68.20", "216.146.25.173", "103.137.63.39", "103.227.224.189", "104.194.152.173", "66.42.52.137", "153.75.94.148", "45.59.124.164", "104.194.153.161", "153.75.95.27", "216.73.158.117", "154.205.154.209", "43.134.62.111", "38.54.17.113", "38.52.17.113"]) | indicator := RemoteAddressIP4 ;
#event_simpleName=DnsRequest | DomainName=/(^|\.)(buildersemp\.sharepoint\.com|cskautogo\.sharepoint\.com|upload\.gofile\.io|twnic\.top|ekmt\.tw)$/i | indicator := lower(DomainName) }
| groupBy([aid, indicator], function=[count(as=events), collect([ContextProcessId]), min(@timestamp, as=first_seen), max(@timestamp, as=last_seen)])Microsoft KQL
let ips = dynamic(["27.102.66.162", "104.194.154.28", "216.73.158.72", "122.147.252.16", "103.106.230.33", "123.58.205.105", "223.165.4.203", "172.86.66.19", "103.214.68.20", "216.146.25.173", "103.137.63.39", "103.227.224.189", "104.194.152.173", "66.42.52.137", "153.75.94.148", "45.59.124.164", "104.194.153.161", "153.75.95.27", "216.73.158.117", "154.205.154.209", "43.134.62.111", "38.54.17.113", "38.52.17.113"]);
DeviceNetworkEvents
| where Timestamp >= datetime(2026-01-01T00:00:00Z)
| extend Host = tolower(tostring(split(tostring(split(replace_regex(RemoteUrl, @"^[a-zA-Z]+://", ""), "/")[0]), ":")[0]))
| extend RIP = replace_string(RemoteIP, "::ffff:", "")
| extend Indicator = case(RIP in (ips), RIP, Host matches regex @"(^|\.)(buildersemp\.sharepoint\.com|cskautogo\.sharepoint\.com|upload\.gofile\.io|twnic\.top|ekmt\.tw)$", Host, "")
| where isnotempty(Indicator)
| summarize Events = count(), Processes = make_set(InitiatingProcessFolderPath, 20), FirstSeen = min(Timestamp), LastSeen = max(Timestamp) by DeviceName, IndicatorIndicators from Kaspersky
Type | Indicator | Kaspersky description |
MD5 | 74b025084026d8d9a64661bf995d835b | MeshAgent |
MD5 | 32151547d912488ac5242731057170a3 | Type 1 Loader |
MD5 | e9d4081fac4a5111e6856f333d490439 | Type 1 Loader |
MD5 | e63ec733393e4e7e3424cbff918003a3 | Type 2 Loader |
MD5 | 3896d44bc6d517291c60ee3f70543a31 | Encrypted Payload (config.ini) |
MD5 | b930fb3ab14fe45b378d5acfa6e0a763 | Encrypted Payload (config.ini) |
MD5 | 6a54f8008ee659b6923abf5dba2b2cc1 | Encrypted Payload (config.ini) |
MD5 | 552f0a9cc5fc3755721fa49be05e8035 | GitDoor |
MD5 | 14ec33ea808e697870d00c5e434c2952 | GraphDoor |
MD5 | d89b2fe0759df2d534e7a728700a6f8c | Modified Erfrp proxy, Go binary (channel.ts) |
MD5 | b4956b230d296cb4b015b42b11d80e32 | Ably channel Go program (abchannel.ts) |
MD5 | 8b78ef5609d942036e93721d75bd1f29 | Legitimate executable, javaw.exe (sideload host, not malware) |
MD5 | be739a7036dc8e2ee8608fba13c8c3a4 | Legitimate executable, start.exe (sideload host, not malware) |
MD5 | 84802caf2579cfe7a38f079a64a3cd22 | Legitimate executable, pythonw.exe (sideload host, not malware) |
MD5 | d955d03104f988582cbd2967177693d3 | Legitimate executable, zoomit.exe (sideload host, not malware) |
MD5 | 86a46af43e18288e1df439b2a9a6f7ca | Attribution table: FamousSparrow overlap, excint.exe (renamed Avast component) |
MD5 | 733f8002aa9531b25384c0014d15cc1f | Attribution table: FamousSparrow overlap, wtsapi32.dll |
MD5 | 91bee19d31d32346cd4d1febb4308606 | Attribution table: GitDoor staging, WINMM.dll |
MD5 | 45b3aa39bc33528650d9181cd946b2df | Attribution table: CopperTurtle overlap, pdfset.exe |
MD5 | 9a5e0ab7466594dcf46943c413db7b81 | Attribution table: GitDoor staging, SbieDll.dll |
IPv4 | 27[.]102[.]66[.]162 | MeshAgent C2 |
IPv4 | 104[.]194[.]154[.]28 | Proxy C2 server |
IPv4 | 216[.]73[.]158[.]72 | Proxy C2 server |
IPv4 | 122[.]147[.]252[.]16 | Proxy C2 server |
IPv4 | 103[.]106[.]230[.]33 | Proxy C2 server |
IPv4 | 123[.]58[.]205[.]105 | Proxy C2 server |
IPv4 | 223[.]165[.]4[.]203 | Proxy C2 server |
IPv4 | 172[.]86[.]66[.]19 | Proxy C2 server |
IPv4 | 103[.]214[.]68[.]20 | Proxy C2 server |
IPv4 | 216[.]146[.]25[.]173 | Proxy C2 server |
IPv4 | 103[.]137[.]63[.]39 | Proxy C2 server |
IPv4 | 103[.]227[.]224[.]189 | Proxy C2 server |
IPv4 | 104[.]194[.]152[.]173 | Proxy C2 server |
IPv4 | 66[.]42[.]52[.]137 | Proxy C2 server |
IPv4 | 153[.]75[.]94[.]148 | Proxy C2 server |
IPv4 | 45[.]59[.]124[.]164 | Proxy C2 server |
IPv4 | 104[.]194[.]153[.]161 | Proxy C2 server |
IPv4 | 153[.]75[.]95[.]27 | Proxy C2 server |
IPv4 | 216[.]73[.]158[.]117 | Proxy C2 server |
IPv4 | 154[.]205[.]154[.]209 | Proxy C2 server |
IPv4 | 43[.]134[.]62[.]111 | Proxy C2 server |
IPv4 | 38[.]54[.]17[.]113 | Proxy C2 server |
IPv4 | 38[.]52[.]17[.]113 | Proxy C2 server |
Domain | upload.gofile[.]io | Legitimate cloud storage |
Domain | buildersemp[.]sharepoint[.]com | SharePoint site |
Domain | cskautogo[.]sharepoint[.]com | SharePoint site |
Domain | twnic[.]top | Cobalt Strike domain |
Domain | ekmt[.]tw | Cobalt Strike domain |
URL | hxxps://github[.]com/LatonyCharnleyaW | GitHub repositories |
URL | hxxps://github[.]com/StevenDFournier | GitHub repositories |
URL | hxxps://github[.]com/VenoraKlanimU | GitHub repositories |
URL | hxxps://github[.]com/JeneLeo | GitHub repositories |
URL | hxxps://github[.]com/AlexandaJazzmynwC | GitHub repositories |
URL | hxxps://github[.]com/FloraineSuvitC | GitHub repositories |
URL | hxxps://github[.]com/bvgarciajocelynxccZ | GitHub repositories |
URL | hxxps://github[.]com/Eclemensadolph1n | GitHub repositories |
URL | hxxps://github[.]com/aYulysseserinYtas | GitHub repositories |
Reading the results before you escalate
Test 1 measures exposure, not compromise.
In our judgment, a Hunt 1 hit on an unapproved host, or Hunt 2 and Hunt 3 on one machine, means isolating it and preserving evidence before anyone uninstalls anything. A verified Hunt 5 match on a ShadowCloud hash is close to proof; a Hunt 6 hit needs host evidence. Hunt 4 alone is a lead until Hunt 1 or Hunt 2 hits the same host. Also check by hand for a C:\ProgramData\Intel_Log folder and a log.lock file in roaming AppData, both artifacts Kaspersky describes.
One next step
If your clients run on Microsoft 365, start with a free Inception Foresight M365 Assessment.



