HTTP/3 RATs target Southeast Asian government networks

src-121d39ef

A cyberespionage campaign targeting government organizations in Southeast Asia has used two custom remote access Trojans (RATs) that communicate with command-and-control infrastructure over HTTP/3, according to an investigation of activity dating back to at least July 2025. The malware, PerfectPioneer and its successor KUKURAT, combine data collection capabilities with techniques intended to make detection and forensic analysis more difficult.

The activity is tracked as TGR-STA-0049 and is also associated with the names REF7707, Earth Alux and Ink Dragon. The group has been active since at least March 2023 and has previously targeted government entities in Southeast Asia, South America and Europe. Earlier operations associated with the group included the Squidoor backdoor.

Multiple routes into targeted organizations

Server exploitation was the primary initial access method observed in the recent campaign. Attackers exploited vulnerable, unpatched Microsoft Exchange servers and web servers, then used web shells to execute commands and deploy malware. The observed web shells included custom tools and publicly available variants such as ASPXSpy, Godzilla, Bysunue.aspx and Suo5.

Phishing was also used as a secondary route. Lures included decoy PDF documents and lookalike domains tailored to targeted organizations. Some malicious executable files were presented with PDF icons and opened a decoy document when launched, while the malware ran in the background. Investigators also observed decoy installers and DLL side-loading in the delivery and execution of PerfectPioneer.

In one example, a decoy installer displayed a legitimate GlobalProtect setup process while deploying PerfectPioneer in the background. This use of legitimate software as cover illustrates why an expected-looking installer or document should not, on its own, be treated as evidence that a process is trustworthy.

HTTP/3 complicates network visibility

PerfectPioneer and KUKURAT use HTTP/3 over QUIC as a primary means of communicating with their command-and-control servers. HTTP/3 carries traffic over UDP, unlike HTTP/2’s TCP transport. The investigation notes that some enterprise firewalls and intrusion detection or prevention systems may not inspect UDP traffic as thoroughly as traditional TCP-based web traffic. Where monitoring is focused on TCP, outbound UDP traffic on port 443 may therefore receive less scrutiny.

PerfectPioneer can fall back to HTTP/2 if its HTTP/3 connection fails. Its communication includes endpoints for agent registration, periodic heartbeats, task retrieval and reporting results. For defenders, this makes protocol-aware monitoring important: allowing UDP 443 may be operationally necessary, but organizations should understand which systems use it and assess whether the traffic is inspected and logged.

The reported behavior also has implications for SIEM and endpoint analytics. Detection strategies that rely primarily on known file signatures or conventional TCP web traffic may miss activity that uses QUIC or executes payloads in memory. Correlating unusual outbound UDP 443 connections with endpoint process activity, web-shell execution, suspicious server changes and authentication events can provide additional context. These are defensive considerations based on the techniques described, not confirmed indicators for any specific organization.

From a feature-rich RAT to a modular framework

PerfectPioneer is written in Go, and investigators identified six variants that had evolved since at least July 2025. The variants differ in their communications and capabilities. A representative version can collect files and system information, run commands, manage processes, capture screenshots, record keystrokes and provide network tunneling functions.

Some versions also contain a routine called isDangerousCommand. Before carrying out an instruction from its operators, the malware checks it against a hard-coded list of commands and blocks selected actions that could be destructive or likely to attract security attention. This mechanism appears designed to reduce operational mistakes that might expose the intrusion.

KUKURAT appeared in compromised networks in March 2026. Written in C++, it is modular: its core can retrieve and load capability modules from the command-and-control server when needed. At least 12 named modules are supported. Two variants were identified, one communicating over HTTP/2 and the other over HTTP/3. The malware also includes configurable mechanisms for generating synthetic network, registry and file activity, which can add benign-looking events around its operations.

Priorities for monitoring and response

Organizations should prioritize patching internet-facing Exchange and web servers, reviewing web-shell exposure and investigating unexpected changes to server files or processes. Security teams can also review outbound QUIC and UDP 443 activity, especially when it is inconsistent with an endpoint’s role or accompanies unusual process execution.

Because the campaign combines phishing, server exploitation, staged payloads and changing malware capabilities, no single alert is likely to provide the full picture. SIEM teams should correlate network, endpoint, identity and server telemetry, while retaining enough detail to investigate suspicious connections and process chains. The investigation describes rapid development of PerfectPioneer variants and a later shift to KUKURAT, underscoring the value of behavior-based detection alongside signatures.