After compromising a government domain controller in Southeast Asia, a threat actor planted malicious log-on scripts and turned the domain's sign-in process into a delivery system. Each time a legitimate user signed in, a script ran on that user's machine and installed a backdoor named KUKURAT. The domain controller did exactly what an authenticated administrator told it to do. In multiple intrusions, a watchdog kept KUKURAT alive, and on one occasion it ran from a folder belonging to Palo Alto Networks' own endpoint security agent.
The threat actor's own tools are far less obedient. Two versions of PerfectPioneer, an earlier backdoor from the same campaign, check every instruction from their controller against a list of forbidden commands before running it. They refuse to wipe a disk, shut down the machine, or reset a firewall, even when the person controlling them gives the order. Unit 42, the threat intelligence team at Palo Alto Networks, documented both backdoors on November 16, 2026, in an analysis of attacks on governments in Southeast Asia.
Unit 42 attributes both backdoors to TGR-STA-0049 (aka Ink Dragon), a China-nexus espionage threat actor it has tracked since at least March 2023. Its targets span government, defense, telecommunications, education, and aviation. The campaign behind the two backdoors has run since at least July 2025.
TGR-STA-0049 wrote distrust into its own malware, then exploited the absence of distrust in its victims' networks. Domain controllers and Exchange servers accepted valid commands without a second check, and the defenders' own security software lent the intrusion a hiding place.
Operation Timeline
Trace the events, turning points, and connections behind the analysis.
TGR-STA-0049 Timeline
Activity and public disclosures tied to TGR-STA-0049 and the overlapping REF7707, Earth Alux, and Ink Dragon activity, from March 2023 to Unit 42's November 2026 analysis. The early entries rest on vendor overlap assessments.
View Timeline Hide Timeline
-
March 2023
-
April to June 2023
-
May 2023
-
Mid-2024
-
November 2024
-
February 13, 2025
-
February 27, 2025
-
March 31, 2025
-
July 2025
-
July 2025
-
December 16, 2025
-
March 2026
-
November 16, 2026
A Backdoor That Distrusts Its User
PerfectPioneer is a remote access trojan written in Go and first observed in July 2025. Unit 42 recovered six variants. RatAAkk-tt, which Unit 42 describes as one of the earliest iterations, speaks only HTTP/2 and lacks persistence. The most complete, newhttp3, adds HTTP/3 and process injection. Across the family, capabilities include keylogging, screenshots, and a SOCKS5 proxy. Two others pose as PDF files, and one of those is stripped down for initial access. The progression reads like a product roadmap.
Two variants, Pioneer and rat_http3_cc, carry a function named isDangerousCommand. It checks every incoming instruction against three layers before execution. The first layer blocks 41 keywords across nine categories. They cover disk destruction, recursive deletion, power actions, and fork bombs. Others target permission changes, account management, service disruption, firewall changes, and boot configuration. The second applies 16 execution-pattern filters against download-and-execute shortcuts and encoded PowerShell. The third protects 12 operating system paths from file deletion.
The blocklist mixes Windows and Linux syntax. kill -9 1 and iptables -F mean nothing on Windows, and Go compiles readily for both platforms, so Unit 42 reads the list as strong evidence of a Linux build in development. The design choice matters more than the platform. The developer wrote the restriction around consequences rather than around an operating system.
Malware routinely defends itself against sandboxes, analysts, and security products. PerfectPioneer also defends the intrusion against its user. Legitimate red team tools restrict themselves too. Cobalt Strike 4.8 added payload guardrails limiting where its Beacon payload runs by IP address, user name, server name, or domain, so a payload executes only on the systems it was built for. PerfectPioneer's rules work after compromise instead, protecting continued access, the infected machine, and the secrecy of the intrusion.
The developer has reason to distrust the people running the intrusion. Elastic Security Labs found the same threat actor, tracked there as REF7707, inside a South American foreign ministry in late November 2024. The intrusion used a renamed copy of Microsoft's signed Console Debugger (CDB), saved as fontdrvhost.exe, to load the FINALDRAFT backdoor. A scheduled task named EPolicyManager relaunched it every minute as SYSTEM. Elastic also found non-functional decryption routines, debug strings, and pre-production samples, and described "poor campaign management and inconsistent evasion practices."
KUKURAT Builds Its Own Alibi
KUKURAT, written in C++, keeps a small core on the infected machine and pulls most capabilities from its command-and-control (C2) server as separate modules. Unit 42 counted support for at least 12. Named modules include BofLoader.dll for Beacon Object Files, SmbBrute.dll for Server Message Block (SMB) password guessing, yuze_dll.dll for SOCKS tunneling, ComExec.dll for execution through the Component Object Model (COM), and MyPsExec.dll for lateral movement. Less code sits on disk, and a module can arrive, run in memory, and disappear.
The HTTP/3 variant keeps one capability in the core, an exec command able to start a process without fetching any module. Unit 42 calls it a built-in survival option for when module delivery fails. One external module, PsExecService.exe, registers a Windows service named PsExec and opens the named pipe \\.\pipe\PSEXEC. It supports shell commands, uploads, downloads, and directory listings. After five minutes without activity, the service uninstalls itself and removes its artifacts.
KUKURAT also manufactures activity with no purpose beyond cover. Each cycle sends ten decoy requests to sites drawn from a list of 15 popular domains. It writes random data to HKCU\Software\Google\Chrome\BrowserCrashRecovery, reads it back, and deletes it. It creates and deletes files with random names and contents under %temp%. The threat actor can switch each stream on or off from the C2 server.
Unit 42 says the noise may significantly affect automated detection and forensic analysis alike. Each event resembles something benign, and each still costs a defender time to resolve. The same noise also gives the malware away. A process with no business touching Chrome's crash recovery key, firing bursts of meaningless web requests, and churning temporary files is one coordinated behavior. Correlating process, account, and host across those three surfaces turns camouflage into a signature.
A companion watchdog keeps KUKURAT alive. The backdoor writes its path to the named pipe WINDOW_SEC_WATCHDOG every three seconds. The watchdog checks every second and relaunches KUKURAT after 15 seconds of silence. On one occasion, the threat actor placed the watchdog in C:\ProgramData\Cyvera\Ransomware\.
Hiding Inside the Defender's Own Stack
C:\ProgramData\Cyvera\ is the working folder of the Cortex XDR agent, Palo Alto Networks' endpoint security product, named for Cyvera, the Israeli company Palo Alto Networks acquired in 2014. The Ransomware subfolder belongs to the agent's anti-ransomware protection. On that occasion, KUKURAT's watchdog hid in the folder of the product whose alerts appear throughout Unit 42's figures, built by the parent company of the team that found it. Unit 42 calls it an attempt to lower suspicion by hiding within security product directories.
Two more Palo Alto Networks products appear in the intrusion. One decoy installer ran a setup program for GlobalProtect, Palo Alto Networks' virtual private network (VPN) client, while PerfectPioneer started in the background. The C2 traffic runs over HTTP/3 on QUIC, a transport Palo Alto Networks has advised its firewall customers to block since 2018 to preserve the firewall's visibility. PerfectPioneer survives that control. When an HTTP/3 connection fails, it falls back to HTTP/2, keeping its C2 channel alive where QUIC is blocked.
The threat actor may have studied its victims' security stack before the intrusion and built camouflage for it, which implies reconnaissance well beyond the network perimeter. An unnamed China-nexus adversary has also exploited PAN-OS firewalls through PHP2Shell. The choices may be generic, since GlobalProtect and Cortex XDR are common in government networks and a Cortex folder is a convincing place to hide anywhere the agent runs. Or the overlap may say as much about the observer as the observed.
Unit 42 sees intrusions through Palo Alto Networks telemetry, and its own figures show Cortex XDR alerts and process trees. Victims running Palo Alto Networks products are the victims Unit 42 is best placed to find. "Southeast Asian governments" describes where this threat actor operates and where this vendor has visibility, and the two maps are not the same. Other vendors see TGR-STA-0049 elsewhere, which is the strongest check on that bias.
Exchange Opens the Archive
Unit 42 names exploitation of unpatched, public-facing Microsoft Exchange and web servers as the primary route in, without naming a vulnerability. The threat actor planted the ASPXSpy and Godzilla web shells, plus Bysunue.aspx and the Suo5 tunneling tool, to keep access after the first exploit. Phishing supplied a second route through executables disguised as PDF documents. One decoy reproduced a 2025 election notice from Myanmar's Union Election Commission.
Other vendors document routes this threat actor used elsewhere. Check Point Research found Ink Dragon forging ASP.NET ViewState payloads signed with publicly disclosed machineKey values, and mass-scanning for the ToolShell SharePoint exploit chain during its first exploitation wave in July 2025. Neither route can be assigned to Unit 42's unnamed victims without evidence, and Unit 42 offers none.
Inside, the threat actor went after the systems holding Windows credentials. It dumped memory from the Local Security Authority Subsystem Service (LSASS) and used DCSync to request password hashes from Active Directory as if it were a domain controller. It ran SharpDPAPI against secrets protected by the Data Protection API (DPAPI) and copied the Security Account Manager (SAM) and SYSTEM registry hives.
One command called comsvcs.dll through rundll32.exe by the ordinal #+000024 instead of the name MiniDump, writing LSASS memory to \Windows\Temp\MxA.tar. The ordinal sidesteps detections searching for the literal string.
A compromised domain controller then turned the victims' own administration into the delivery system. A malicious domain log-on script installed KUKURAT whenever a legitimate user signed in, and one alert caught AdobePhoto.exe running at startup. It is the same inversion I examined when VOODOO BEAR (aka Sandworm) wiped a grain firm through its own Active Directory, put to quieter use.
Next, the threat actor searched for accounts holding ApplicationImpersonation, a role letting one service account act as any mailbox owner. The commands loaded the on-premises Exchange management snap-in. Microsoft completed removal of ApplicationImpersonation from Exchange Online in March 2025. Governments keeping mail on their own servers kept the role and the exposure with it. VENOMOUS BEAR (aka Turla) has also run espionage tooling from victims' own Exchange servers, where the mail archive and its permissions sit together.
Where impersonation was unavailable, the threat actor built its own access. It ran Get-Mailbox -ResultSize Unlimited and passed every result to Add-MailboxPermission, granting one compromised account FullAccess to every mailbox with auto-mapping disabled, so Outlook would not add the mailboxes to the account's profile automatically. Unit 42 saw attempts to steal PDF, Word, Excel, and image files whose names pointed to international relations and internal political affairs. As in Denmark, legitimate permissions did the work.
Who Wrote the Rules
The FullAccess loop is the loudest action in the intrusion. Exchange logs permission changes, and mailbox auditing records delegated access. A developer who forbids iptables -F for fear of exposure sits in the same campaign as someone granting one account every mailbox in a government agency.
Hypothesis 1, Separate Developers and Intrusion Teams
A development function writes the tools and encodes its distrust of the people using them. A separate intrusion team works to its own collection deadlines and accepts risks the developers would not. It is the same shared development function I weighed in CopperDragon. Elastic's record of sloppy campaign management alongside capable engineering supports the split, and so do six PerfectPioneer variants tuned to stages of an intrusion and a modular successor built to accept capabilities Unit 42 has never seen.
Hypothesis 2, Espionage Rules of Engagement
The blocklist keeps the intrusion on the espionage side of a political line. Destroying disks, crashing servers, and disabling firewalls inside a government agency converts quiet collection into an attack a government must answer. Exposed espionage carries a diplomatic cost Beijing has repeatedly absorbed, since every major state collects. Destruction invites a different response. Compare the prepositioning the Cybersecurity and Infrastructure Security Agency (CISA) described for VANGUARD PANDA (aka Volt Typhoon) inside American critical infrastructure, built for disruption. PerfectPioneer's blocklist forbids disruption outright.
Hypothesis 3, Insurance Against a Seized C2 Server
Someone else may one day control the C2 channel. In January 2025, the Federal Bureau of Investigation (FBI) sent commands through PlugX's own infrastructure to remove it from about 4,258 American computers, after Sekoia took over a PlugX C2 server. On November 13, the Bundestag voted to let Germany's foreign intelligence service, in narrow cases, shut down servers run by state-backed threat actors. A hard-coded blocklist limits what a foreign service could do with a seized PerfectPioneer channel, including a destructive act blamed on Beijing.
The counterargument is strong. The blocklist is a command filter, not a security boundary. A service holding the server could use alternate syntax, upload its own tooling, or reach capabilities the filter never checks. The safeguard raises the cost of misuse without preventing it, which makes it a weak insurance policy against a capable seizure.
Hypothesis 4, A Contractor Protecting Its Product
China's offensive ecosystem runs partly on outside suppliers. MI5 says the Ministry of State Security funded UK-linked academics through an intermediary to outsource research, and Washington sanctioned Sichuan Juxinhe over the Salt Typhoon telecom intrusions. The Justice Department's March 2025 indictment of i-Soon described sales to the Ministry of Public Security and Ministry of State Security at an estimated US$10,000 to US$75,000 per exploited inbox, with Asian foreign ministries among the targets.
In that market, persistent access is the product, and a tool vendor has every reason to stop a buyer's intrusion team from destroying it. The same market explains the FullAccess loop. A contractor paid by the inbox gains from taking every inbox at once.
Hypothesis 5, Deconfliction With Other Chinese Intrusions
Check Point found RudePanda, a second China-nexus intrusion set, inside several of the same government networks and assessed the two as not operationally linked. A destructive command from one threat actor would also burn the access of another service collecting from the same agency. Rules preventing collateral damage protect the national collection effort, not only one intrusion.
The evidence supports the first hypothesis most strongly, because it explains both the blocklist and the loud mailbox grab. The second fits the blocklist's categories, which target destruction and disruption specifically while leaving theft and surveillance untouched. The third and fifth fit the code without any direct evidence for them. The fourth explains the commercial logic of both the restriction and the mailbox loop, but nothing in the public record identifies a contractor.
Collection Requirements and Their Limits
The filenames point to international relations and internal politics, and the region supplies obvious questions. At their November 15 summit in Manila, the Association of Southeast Asian Nations (ASEAN) and China again failed to conclude a South China Sea code of conduct. About 80% of the draft is tentatively agreed, and whether the code binds its parties remains unresolved. A ministry's mail can expose its negotiating position, its internal divisions, and the concessions it would accept.
Geopolitical fit cannot establish a specific tasking. "International relations" covers bilateral diplomacy, trade and infrastructure talks, relations with Washington, and elite politics as readily as the South China Sea. No victim, document, or collection order has been named. A link to the code of conduct talks is plausible and unproven.
The Myanmar lure points to another requirement. The military government held voting from December 2025 through January 2026 in 263 of 330 townships, with opposition parties banned. Volker Türk, the United Nations High Commissioner for Human Rights, said the process "served only to exacerbate violence and societal polarisation," according to UN News. Chinese Foreign Minister Wang Yi had voiced support for the vote, The Irrawaddy reported. An election notice would interest officials across the region, and Unit 42 does not identify the recipient.
The infrastructure suggests a wider target set than Unit 42's title. KUKURAT's C2 domains include www.police.gr[.]com, imitating the Greek police, and www.investigationj6[.]com. PerfectPioneer used www.thailisting[.]net and www.yangon.it[.]com. Check Point documented European government victims, and Elastic's foreign ministry sits in South America. Domain names are lures and labels, not victim lists, but they point well beyond Southeast Asia.
Ink Dragon turns compromised servers into a relay network, and Check Point found "multiple instances where government entities were inadvertently serving as C2 relays," with European victims used to reach targets in Africa and Southeast Asia. One government's server can deliver another government's intrusion. Traffic arriving from a friendly ministry carries no attribution value, and a victim can become the apparent source of an attack on its neighbor.
Arguing the Other Side
Beijing routinely rejects such accusations. When American authorities seized hacking infrastructure in August 2026, Foreign Ministry spokesperson Lin Jian said China would "not accept any vilification in the name of so-called cybersecurity issues" and pointed to an alleged National Security Agency attack on China's National Time Service Center, Outlook India reported. I found no Chinese response to Unit 42's analysis as of November 18. The counter-accusation holds one point worth conceding. China is not the only state collecting against Southeast Asian governments.
The attribution rests on vendors rather than governments. Unit 42 promoted this activity from an activity cluster, CL-STA-0049, to a temporary threat group, a step its attribution framework reserves for activity confidently tied to one distinct actor not yet linked to a named one. In February 2025, Unit 42 assessed the activity originated in China, based on tactics and techniques. The aliases are vendor judgments of overlap, and each vendor's view is partial.
I found no public confirmation from an affected government as of November 18, and none has named China. Southeast Asian states rarely attribute cyber espionage to Beijing publicly, and ASEAN is China's largest trading partner. In that silence, a commercial vendor's report becomes the only public record, written by a company whose products feature in the intrusion and are recommended in Unit 42's closing section.
A criminal explanation fits the capabilities better than the victims. Credentials, mailbox access, tunnels, and stolen documents all sell in the eCrime ecosystem. Repeated targeting of foreign ministries and other government bodies, the absence of extortion or resale, and access held for months point to intelligence collection instead.
New backdoors have followed public exposure within months. Elastic, Unit 42, and Trend Micro published on this activity between February and March 2025, and PerfectPioneer appeared in July 2025. Check Point exposed the relay network in December 2025, and KUKURAT arrived in March 2026. The sequence fits a threat actor retooling after each disclosure, though PerfectPioneer's development may predate the 2025 reports, so the pattern shows timing rather than cause.
The blocklist, protocol fallback, noise generator, watchdog, and modular design show sustained investment in staying hidden. The FullAccess loop shows an intrusion team willing to spend that stealth on collection. Trend Micro tracks the same activity as Earth Alux. Together with Unit 42, Elastic, and Check Point, it documents attacks on government targets since 2023. They span Southeast Asia, South America, and Europe.
I assess with high confidence the campaign serves state intelligence collection rather than financially motivated crime. I assess with moderate confidence TGR-STA-0049 serves a Chinese state requirement and separates at least part of its malware development from hands-on intrusion work. I hold any specific tasking and any contractor role at low confidence. No victim, document, or employer has been identified. Attribution remains a confidence game, and the public evidence identifies engineers without identifying who employs them.
Valid Instructions, No Second Check
PerfectPioneer and KUKURAT will not stay stable detection targets. Their developers have already moved from an all-in-one Go backdoor to a modular C++ framework, changed transports, and hashed their endpoint names. KUKURAT shares /register and /heartbeat with PerfectPioneer's /agent/ routes but hashes them, turning /heartbeat into /c8a4e192.
Hashes and filenames remain useful for known samples, and Unit 42 publishes them in full. The behaviors needed to complete the mission last longer, because a new module or transport does not change what the threat actor must do inside an Exchange server or a domain controller.
- Exchange permissions. Audit ApplicationImpersonation assignments and alert on
Add-MailboxPermissiongranting FullAccess across many mailboxes, especially with auto-mapping disabled. - Domain log-on scripts. Review scripts in SYSVOL and Group Policy for unexpected executables such as
AdobePhoto.exe. - Credential dumping. Alert on
rundll32.exeloadingcomsvcs.dll, including calls by ordinal. - QUIC. Restrict outbound User Datagram Protocol (UDP) port 443 to systems with a documented need, and inspect the HTTP/2 traffic a block forces back onto the Transmission Control Protocol (TCP).
- Security agent folders. Alert on executables writing to or launching from
C:\ProgramData\Cyvera\other than the Cortex XDR agent itself, and treat any process in a security product's folder as suspect until verified.
PerfectPioneer's developer assumed a valid instruction from an authenticated user could still be dangerous, and wrote a second check into the code. The governments around it made the opposite assumption.
A domain controller pushed KUKURAT to every user who signed in, because a log-on script told it to. Exchange opened every mailbox in an agency to one account, because an administrator told it to. Neither system asked whether the action still served the organization issuing it. The malware held a veto over the person using it, and the victims held none over theirs.
Analyst Annex
Evidence status as of November 18, 2026. The records cover the TGR-STA-0049 campaign Unit 42 documented against government entities in Southeast Asia, active since at least July 2025, including the PerfectPioneer and KUKURAT backdoors, credential theft, domain log-on script delivery, and Exchange mailbox access. Unit 42 names no victim country or organization.
Earlier REF7707, Ink Dragon, Earth Alux, and Squidoor activity, RudePanda’s presence in the same networks, and the ToolShell scanning wave appear under Related Activity as context only. None of those procedures is mapped to Unit 42’s victims.
Confirmed marks evidence published by a victim, vendor, regulator, or government authority about its own systems, decisions, or investigation. Reported marks technical findings or attribution assessments published by a third party. Claimed marks statements by a seller or self-identified threat actor that remain unverified. Provisional marks an analytical mapping supported by reported behavior but not made by the original source.
Indicators
The table lists every network indicator Unit 42 published and eight host and network artifacts. Unit 42 also publishes SHA256 hashes for the PerfectPioneer variants, loader DLLs, five KUKURAT samples, the watchdog, the PsExecService.exe module, SharpDPAPI builds, 13 web shells, and five unmapped files. The full Unit 42 indicator list holds them.
| Indicator | Context | Status |
|---|---|---|
| IP Addresses | ||
27.254.164[.]212 |
IP address Unit 42 lists for the campaign Source: Unit 42 | Reported |
27.254.164[.]209 |
IP address Unit 42 lists for the campaign, in the same range as the address above Source: Unit 42 | Reported |
| Domains | ||
www.yangon.it[.]com |
PerfectPioneer command-and-control (C2) domain referencing Myanmar’s largest city Source: Unit 42 | Reported |
www.thailisting[.]net |
PerfectPioneer C2 domain referencing Thailand Source: Unit 42 | Reported |
www.maui-cocktailbar[.]com |
PerfectPioneer C2 domain Source: Unit 42 | Reported |
ads-edge[.]com |
PerfectPioneer C2 domain Source: Unit 42 | Reported |
www.wkuyda[.]com |
PerfectPioneer C2 domain Source: Unit 42 | Reported |
www.cloudflacecdn[.]com |
KUKURAT C2 domain imitating Cloudflare’s content delivery network Source: Unit 42 | Reported |
www.police.gr[.]com |
KUKURAT C2 domain imitating the Greek police Source: Unit 42 | Reported |
www.investigationj6[.]com |
KUKURAT C2 domain Source: Unit 42 | Reported |
www.sacofficegov[.]com |
Additional campaign domain imitating a government name Source: Unit 42 | Reported |
| Host and Network Artifacts | ||
C:\ProgramData\Cyvera\Ransomware\ |
Folder holding the KUKURAT watchdog on one occasion. The Cortex XDR agent uses C:\ProgramData\Cyvera\ as its working folder
Source: Unit 42, Cortex XDR documentation
|
Reported |
WINDOW_SEC_WATCHDOG |
Named pipe KUKURAT writes its path to every three seconds. The watchdog relaunches KUKURAT after 15 seconds of silence Source: Unit 42 | Reported |
\\.\pipe\PSEXEC |
Named pipe opened by the PsExecService.exe module, which registers a Windows service named PsExec and removes itself after five minutes without activity
Source: Unit 42
|
Reported |
\Windows\Temp\MxA.tar |
LSASS memory dump written by rundll32.exe calling comsvcs.dll through the ordinal #+000024
Source: Unit 42
|
Reported |
AdobePhoto.exe |
File an alert caught running at startup after delivery through a malicious domain log-on script Source: Unit 42 | Reported |
/agent/register |
First of four PerfectPioneer C2 routes, alongside /agent/heartbeat, /agent/tasks, and /agent/result
Source: Unit 42
|
Reported |
/c8a4e192 |
Hashed KUKURAT C2 route standing in for /heartbeat
Source: Unit 42
|
Reported |
HKCU\Software\Google\Chrome\BrowserCrashRecovery |
Registry key KUKURAT writes, reads back, and deletes as decoy activity Source: Unit 42 | Reported |
MITRE ATT&CK Mapping
| Tactic | Technique | Procedure | Status |
|---|---|---|---|
| TGR-STA-0049 Campaign Against Southeast Asian Government Entities | |||
| Initial Access | T1190 Exploit Public-Facing Application | The threat actor exploited unpatched, public-facing Microsoft Exchange and web servers. Unit 42 names no vulnerability Source: Unit 42 | Reported |
| Execution | T1204.002 User Execution: Malicious File | Victims ran executables disguised as PDF documents, which opened decoys such as a 2025 Myanmar Union Election Commission notice while the malware ran Source: Unit 42 | Reported |
| Persistence | T1505.003 Server Software Component: Web Shell |
The threat actor planted ASPXSpy, Godzilla, Bysunue.aspx, and Suo5 on compromised servers
Source: Unit 42
|
Reported |
| Persistence | T1037.003 Boot or Logon Initialization Scripts: Network Logon Script | A malicious domain log-on script on a compromised domain controller installed KUKURAT when legitimate users signed in Source: Unit 42 | Reported |
| Persistence | T1098.002 Account Manipulation: Additional Email Delegate Permissions |
A loop piping Get-Mailbox -ResultSize Unlimited into Add-MailboxPermission granted one account FullAccess to every mailbox with auto-mapping disabled
Source: Unit 42
|
Reported |
| Execution | T1569.002 System Services: Service Execution |
The PsExecService.exe module registers and runs a Windows service named PsExec
Source: Unit 42
|
Reported |
| Execution | T1047 Windows Management Instrumentation | The threat actor used Windows Management Instrumentation to move across the network Source: Unit 42 | Reported |
| Stealth | T1036.008 Masquerading: Masquerade File Type | PerfectRat and rat_http3_ls variants posed as PDF documents Source: Unit 42 | Reported |
| Stealth | T1036.005 Masquerading: Match Legitimate Resource Name or Location |
On one occasion, the KUKURAT watchdog ran from the Cortex XDR agent’s Cyvera\Ransomware folder, and a decoy installer ran a GlobalProtect setup program while PerfectPioneer started
Source: Unit 42, Cortex XDR documentation
|
Provisional |
| Stealth | T1218.011 System Binary Proxy Execution: Rundll32 |
rundll32.exe called comsvcs.dll by the ordinal #+000024 instead of the name MiniDump
Source: Unit 42
|
Reported |
| Stealth | T1055 Process Injection |
The newhttp3 PerfectPioneer variant injects code through VirtualAllocEx, WriteProcessMemory, and CreateRemoteThread
Source: Unit 42
|
Reported |
| Stealth | T1070.004 Indicator Removal: File Deletion |
The PsExecService.exe module uninstalls itself and removes its artifacts after five minutes without activity, and PerfectPioneer removes its temporary files
Source: Unit 42
|
Reported |
| Credential Access | T1003.001 OS Credential Dumping: LSASS Memory |
LSASS memory was dumped to \Windows\Temp\MxA.tar, and Unit 42 also lists Mimikatz and ProcDump
Source: Unit 42
|
Reported |
| Credential Access | T1003.006 OS Credential Dumping: DCSync | The threat actor requested password hashes from Active Directory through DCSync Source: Unit 42 | Reported |
| Credential Access | T1003.002 OS Credential Dumping: Security Account Manager | The threat actor copied the SAM and SYSTEM registry hives Source: Unit 42 | Reported |
| Credential Access | T1555 Credentials from Password Stores | SharpDPAPI decrypted secrets protected by the Data Protection API Source: Unit 42 | Provisional |
| Credential Access | T1110 Brute Force |
The KUKURAT SmbBrute.dll module guesses SMB credentials
Source: Unit 42
|
Reported |
| Credential Access | T1056.001 Input Capture: Keylogging | PerfectPioneer and a KUKURAT module record keystrokes Source: Unit 42 | Reported |
| Discovery | T1069 Permission Groups Discovery |
Get-ManagementRoleAssignment -Role:ApplicationImpersonation searched for accounts able to impersonate any mailbox owner
Source: Unit 42
|
Provisional |
| Discovery | T1046 Network Service Discovery | KUKURAT port scan and host scan modules mapped the network Source: Unit 42 | Reported |
| Lateral Movement | T1021.002 Remote Services: SMB/Windows Admin Shares |
The MyPsExec.dll module and PsExec moved the threat actor between machines
Source: Unit 42
|
Reported |
| Lateral Movement | T1021.006 Remote Services: Windows Remote Management | The threat actor used Windows Remote Management for lateral movement Source: Unit 42 | Reported |
| Collection | T1114.002 Email Collection: Remote Email Collection | FullAccess grants gave one account every mailbox on the on-premises Exchange server Source: Unit 42 | Provisional |
| Collection | T1005 Data from Local System | Unit 42 saw attempts to steal PDF, Word, Excel, and image files with names pointing to international relations and internal political affairs Source: Unit 42 | Reported |
| Collection | T1113 Screen Capture | PerfectPioneer and a KUKURAT module capture screenshots Source: Unit 42 | Reported |
| Command and Control | T1071.001 Application Layer Protocol: Web Protocols | Both backdoors communicate over HTTP/3 on QUIC, and some variants use HTTP/2 Source: Unit 42 | Reported |
| Command and Control | T1008 Fallback Channels | PerfectPioneer falls back from HTTP/3 to HTTP/2 when its primary channel fails, keeping the channel alive where QUIC is blocked, as Palo Alto Networks recommends Source: Unit 42 | Reported |
| Command and Control | T1105 Ingress Tool Transfer | KUKURAT pulls most capabilities from its C2 server as modules, and runs them in memory Source: Unit 42 | Reported |
| Command and Control | T1090 Proxy |
PerfectPioneer offers a SOCKS5 proxy, and the KUKURAT yuze_dll.dll module tunnels SOCKS traffic
Source: Unit 42
|
Reported |
Entities
| Entity | Role | Status |
|---|---|---|
| Threat Actors | ||
| TGR-STA-0049 (aka Ink Dragon)Threat actor, nation state | China-nexus espionage threat actor Unit 42 has tracked since at least March 2023 against government, defense, telecommunications, education, and aviation targets. Other vendors track overlapping activity as REF7707 and Earth Alux Source: Unit 42, Check Point Research | Reported |
| RudePandaThreat actor, nation state | China-nexus intrusion set Check Point found inside several Ink Dragon victim networks. Check Point does not assess the two as operationally linked Source: Check Point Research | Reported |
| Organizations and Victims | ||
| Unnamed Southeast Asian Government EntitiesVictims | Targets of the campaign. Unit 42 names no country, agency, or victim count Source: Unit 42 | Reported |
| Palo Alto NetworksCyber security vendor | Parent company of Unit 42. Its GlobalProtect client and, on one occasion, its Cortex XDR agent folder appear in the intrusion as camouflage, and its telemetry appears in Unit 42’s figures Source: Unit 42, Cortex XDR documentation | Reported |
| Union Election Commission of MyanmarImpersonated body | Its 2025 election notice served as a phishing decoy Source: Unit 42 | Reported |
| Tools and Malware | ||
| PerfectPioneerRemote access trojan |
Go backdoor first observed in July 2025 in six variants. Two carry isDangerousCommand, blocking 41 keywords, 16 execution patterns, and file deletion in 12 protected paths
Source: Unit 42
|
Reported |
| KUKURATRemote access trojan | Modular C++ backdoor observed from March 2026, with at least 12 modules, decoy activity generation, and HTTP/2 and HTTP/3 variants Source: Unit 42 | Reported |
| KUKURAT WatchdogPersistence component |
Deployed in multiple intrusions. Relaunches KUKURAT after 15 seconds without a heartbeat on the WINDOW_SEC_WATCHDOG pipe
Source: Unit 42
|
Reported |
| ASPXSpy, Godzilla, Bysunue.aspx, and Suo5Web shells and tunneling tool | Kept server access after initial exploitation Source: Unit 42 | Reported |
| SharpDPAPICredential theft tool | Decrypted secrets protected by the Data Protection API Source: Unit 42 | Reported |
| Infrastructure and Services | ||
| On-Premises Microsoft ExchangeCollection target | Retains the ApplicationImpersonation role Microsoft removed from Exchange Online in March 2025 Source: Unit 42, Microsoft | Reported |
Related Activity
The records below cover earlier activity other vendors attribute to the same or overlapping threat actor. None was observed in the intrusions Unit 42 documented.
| Item | Detail | Status |
|---|---|---|
| REF7707 Foreign Ministry IntrusionIntrusion, November 2024 |
Elastic found the threat actor in a South American foreign ministry, loading FINALDRAFT through a renamed copy of Microsoft’s Console Debugger relaunched every minute by the EPolicyManager scheduled task, and described poor campaign management and inconsistent evasion practices
Source: Elastic Security Labs
|
Reported |
| Squidoor BackdoorMalware, reported February 2025 | Unit 42 tracked the activity as CL-STA-0049 and documented Squidoor with ten Windows communication methods, including the Outlook mail API, DNS tunneling, and ICMP tunneling Source: Unit 42 | Reported |
| Earth Alux ActivityCampaign, 2023 onward | Trend Micro reported attacks in Thailand, the Philippines, Malaysia, Taiwan, and Brazil, using the VARGEIT and COBEACON backdoors after exploiting internet-facing servers Source: Trend Micro | Reported |
| Ink Dragon Relay NetworkInfrastructure, reported December 2025 | Check Point found compromised IIS servers relaying C2 traffic, including government servers relaying attacks on other governments in Europe, Africa, and Southeast Asia Source: Check Point Research | Reported |
| ViewState and ToolShell ExploitationInitial access, 2025 | Check Point found forged ASP.NET ViewState payloads signed with publicly disclosed machineKey values, and mass scanning for the ToolShell SharePoint exploit chain in July 2025 Source: Check Point Research | Reported |
Intelligence Gaps
| Gap | Why It Matters | What Would Close It |
|---|---|---|
| Victim identities and countries | Without named victims, the collection requirement and any link to the South China Sea talks or the Myanmar election cannot be tested | Victim disclosures, government statements, or regional CERT advisories |
| Initial access vulnerability | Unit 42 names Exchange and web server exploitation without a CVE, leaving defenders unable to confirm exposure on their own servers | Unit 42 vulnerability details, victim forensics, or vendor advisories tied to the campaign |
| Collection outcome | Unit 42 reports attempts to steal documents, not confirmed theft, and does not say whether the FullAccess grants were used to read mail | Mailbox audit logs, exfiltration telemetry, or victim incident reports |
| Structure behind the code | The split between developers and the intrusion team rests on inference from the blocklist and the loud mailbox loop | Recovered tasking, infrastructure ownership, contractor records, or government indictments |
| Reason for the Palo Alto Networks camouflage | The GlobalProtect decoy and Cyvera folder may reflect reconnaissance of victims, generic camouflage, or the vendor’s own visibility | Victim security stack records, other vendors’ samples showing the same folder, or recovered targeting notes |
| Purpose of the blocklist | Control over the intrusion team, espionage rules of engagement, protection against a seized C2 server, product protection, and deconfliction all fit the code | Developer documentation, recovered intrusion team communications, or behavior of later variants |
| Linux variant | Unit 42 reads the blocklist’s Linux syntax as evidence of a Linux build in development, which would extend the campaign to Linux servers | Recovered Linux samples or telemetry from Linux hosts |
| Eviction status | Unit 42 does not say whether the threat actor has been removed from the victim networks | Victim remediation statements or later Unit 42 reporting |