Skip to main content
Emerging ThreatsMalware & Ransomware

Chinese Hackers Deploy HTTP/3 RATs in Southeast Asian Cyberespionage Campaign

Formal individual walks down a Southeast Asian government office hallway with subtle tech infrastructure in the background.

"Unit 42 has uncovered a multi-stage cyberespionage campaign that has been targeting Southeast Asian government entities since at least July 2025," Unit 42 reported, describing an operation that combines novel protocol-level evasion with rigorous endpoint stealth.

TGR-STA-0049: scope, timeline and choice of targets

Unit 42 tracks the activity as TGR-STA-0049 (also REF7707, Earth Alux, Ink Dragon). The group has been active since at least March 2023 and, in the most recent campaign beginning mid-2025, focused on government organizations across Southeast Asia with additional activity noted in South America and Europe. The operators pursued long-term intelligence collection and data exfiltration, repeatedly refining their toolset across a nine-month period that saw an explicit migration from PerfectPioneer RAT to a newer framework named KUKURAT.

Protocol-level evasion: HTTP/3 (QUIC) as a command-and-control channel

Both PerfectPioneer RAT and KUKURAT use HTTP/3 as the primary method for command-and-control communications. Unit 42 highlights that HTTP/3 runs over the QUIC transport protocol on UDP and that outbound UDP port 443 traffic can bypass traditional TCP-focused deep packet inspection. The report states that both families fall back to HTTP/2 when HTTP/3 fails, and that some KUKURAT variants use HTTP/2 while others use HTTP/3, underscoring deliberate experimentation with transport choices to evade detection.

PerfectPioneer RAT: newhttp3 variant and the isDangerousCommand “safety switch”

PerfectPioneer RAT is a Go-based family observed in at least six variants (including newhttp3, PerfectRat, Pioneer, RatAAkk-tt, rat_http3_ls and rat_http3_cc). Unit 42 analyzed the newhttp3 variant in depth and documented a broad native command set: file operations (including chunked transfers), shell execution, process management, code injection, keylogging, SOCKS5 proxying, screenshots and resource monitoring. The agent implements automated reconnection, cleanup and resource-monitoring routines designed to maintain a low forensic footprint.

Two variants (Pioneer and rat_http3_cc) implement an unprecedented isDangerousCommand routine that blocks a hard-coded list of noisy or destructive commands before execution. Unit 42 shows the blocklist covers 41 explicit destructive or administrative keywords and additional heuristic and directory-protection layers that prevent, for example, disk-destruction strings, chained download-and-execute pipelines and deletion of core OS paths on Windows and Linux.

KUKURAT: a modular, memory-resident framework with heuristic camouflage

Introduced in March 2026, KUKURAT is a C++-based, modular framework whose core binary loads many post-exploitation capabilities as DLL modules from the C2 at runtime. Unit 42 identified at least 12 named modules covering screenshots, keylogging, host and port scanning, process management and lateral-movement tools such as PsExec and SMB credential brute-forcing. One recovered KUKURAT companion module is a PsExec-style service (PsExecService.exe) that registers as a Windows service, supports shell/file operations and self-destructs after five minutes of inactivity.

KUKURAT also embeds a three-axis heuristic camouflage engine that generates network noise (decoy HTTP requests to popular sites), mimics Chrome crash-recovery registry activity, and creates short-lived temp files to obscure real behavior. The agent can pair with a watchdog process that relaunches KUKURAT if it is killed, and the operators even placed the watchdog under a Cortex XDR–associated directory to lower suspicion.

Infection vectors, credential theft and targeted exfiltration

Server exploitation—most notably of unpatched Microsoft Exchange servers and other web servers—was the primary initial access method observed, with web shells (including ASPXSpy, Godzilla, Bysunue.aspx and Suo5) used to stage payloads. Unit 42 also documented secondary vectors: phishing with decoy PDFs (one impersonating a 2025 Union Election Commission document for Myanmar) and decoy installers (one example wrote a GlobalProtect installer while dropping the RAT in the background). DLL side-loading and loader DLLs embedding XOR-encrypted shellcode were also used.

After access, the actor moved laterally using domain log-on script modification, PsExec, WMI and WinRM. Credential theft tactics included LSASS memory dumping (including misuse of comsvcs.dll invoked by ordinal #+000024), Mimikatz, ProcDump, SharpDPAPI and DCSync. The group targeted document types (.pdf, .docx, .xlsx, .png) and abused Exchange privileges—searching for ApplicationImpersonation roles and, where needed, running commands that loop through mailboxes to grant FullAccess privileges—enabling broad mailbox reads and likely data exfiltration.

What this means for technologists, policymakers, and affected enterprises

  • Technologists and security teams: Unit 42 urges logging or restricting QUIC/UDP 443, auditing Active Directory log-on scripts and adding behavioral detections for ordinal-based LSASS dumping (rundll32.exe interacting with comsvcs.dll via ordinal #+000024).
  • Policymakers and regulators: The campaign highlights a gap where legacy, TCP-centric inspection policies may miss sophisticated espionage that uses UDP-based web traffic and runtime module loading that evades static scans.
  • Enterprises and procurement leaders: Unit 42 recommends strict auditing of Exchange impersonation rights, monitoring for automated Add-MailboxPermission loops, and considering endpoint and network defenses that capture unauthorized UDP-based web traffic and inspect QUIC flows.

Unit 42’s analysis paints a picture of a patient, evolving espionage program that pairs protocol-level innovation with endpoint-level camouflage. Defenders are left with a clear technical task: extend visibility to UDP/QUIC channels and strengthen host-level behavioral telemetry to match the attacker’s emphasis on staying unseen.

https://unit42.paloaltonetworks.com/custom-rats-target-southeast-asia-governments/