1. Description
BRONZE SANDCASTLE is a threat activity designation used by Sophos researchers to track a cluster of intrusions involving modified Meterpreter-based backdoors and distinctive DLL sideloading behavior.
DLL sideloading occurs when an application loads a malicious dynamic-link library (DLL) in place of, or alongside, an expected library. Attackers can abuse this behavior by placing a malicious DLL where a legitimate executable will load it. This allows malicious code to run within the execution context of a trusted application.
In the activity documented by Sophos, the attackers paired legitimate executable files with malicious DLLs. The malware then accessed encrypted configuration or payload files and executed decoded code in memory.
The resulting backdoor could collect information about the compromised host, communicate with external infrastructure, and support additional post-exploitation activity. Sophos also observed activity involving network enumeration and access to sensitive Active Directory data.
The use of legitimate software, encrypted payloads, and memory-based execution can complicate detection when organizations rely primarily on static file signatures.
2. Technical Description
The reported activity combines the following techniques:
DLL sideloading: Legitimate executable files load attacker-controlled DLLs.
Encrypted payload storage: Payloads or configuration data are stored in files such as
config.ini.Runtime decryption: Malicious components decode the stored content during execution.
Reflective loading: Executable content is loaded into memory.
Control-flow obfuscation: A flattened dispatcher makes the program's execution flow more difficult to analyze.
Modified Meterpreter functionality: The payload resembles Metasploit-associated shellcode but has implementation differences.
COFFLoader-related functionality: Components suggest additional custom post-exploitation capabilities.
System reconnaissance: The backdoor collects host, process, account, and network information.
Command-and-control communication: The payload communicates with external infrastructure and may retrieve encrypted configuration data.
Active Directory access: Researchers observed activity involving Volume Shadow Copies and the
ntds.ditdatabase file.
These behaviors form a multi-stage intrusion pattern. However, not every technique necessarily appears in every incident.
3. Attack Overview
Stage 1: Executable and DLL Preparation
The operators use legitimate executables alongside malicious DLLs. The executable provides a plausible application context, while the DLL introduces the attacker's code.
Sophos repeatedly observed Sandboxie-related executables paired with sbiedll.dll.
Stage 2: DLL Sideloading
When the executable runs, it loads the malicious DLL. The DLL then executes unauthorized code.
Because the executable may be legitimate, defenders must examine the DLL's location, signature, loading relationship, and behavior rather than relying on the executable's name alone.
Stage 3: Encrypted Payload Retrieval
The malicious component reads encrypted data from a file such as config.ini. The payload is decrypted during execution, making the content less obvious to conventional file scanning.
Stage 4: In-Memory Execution
The decoded payload contains a loader capable of loading executable content into memory. This can reduce the need to write the complete executable payload to disk.
Stage 5: Host Profiling
The backdoor collects information about the compromised system, including host identifiers, operating-system details, network addresses, and process metadata.
Stage 6: Command-and-Control Communication
The payload communicates with external infrastructure. Sophos also observed a version of the payload retrieving encrypted configuration files from GitHub.
Stage 7: Reconnaissance and Post-Exploitation
The operators use commands and other functionality to enumerate accounts and network resources. Sophos additionally observed Volume Shadow Copy creation followed by copying of ntds.dit.
This activity can expose sensitive identity information and create opportunities for further compromise.
4. Technical Analysis
4.1 DLL Sideloading Through Legitimate Applications
Sophos identified multiple executable-and-DLL combinations associated with the activity.
| Executable | Associated DLL |
|---|---|
SandboxieWUAU.exe | sbiedll.dll |
Start.exe — renamed Sandboxie executable | sbiedll.dll |
javaw.exe | jli.dll |
Krita.exe | Krita.dll |
NvdaHelper.exe | NvdaHelperRemote.dll |
Jcef_helper.exe | Libcef.dll |
Vmmap.exe | Version.dll |
LineAppMgr.exe | msvcp140.dll |
bginfo64.exe | snmpapi.dll |
The names above are reported in the Sophos research. Some correspond to legitimate software components and should not be treated as malicious solely because their names appear in this table.
Security significance: An otherwise legitimate executable may become part of an attack when it loads an unexpected DLL from an untrusted location. Monitoring DLL-load events and their associated file paths can help identify this behavior.
4.2 Encrypted Configuration Files
Sophos reported that malicious DLLs typically read encrypted payloads from local files, including config.ini.
One analyzed archive, aez.zip, contained:
javaw.exejli.dllconfig.ini
The DLL reads the encrypted content and supports the execution of the decoded payload.
Separating the loader from the encrypted payload may allow operators to change payload data without replacing the loader itself. This can complicate detection based on a single file hash.
Defensive focus: Investigate unexpected DLLs that read local configuration files shortly before suspicious memory execution or external network communications.
4.3 Control-Flow Obfuscation
Researchers identified a flattened control-flow dispatcher in the malicious DLL code.
Instead of following a straightforward sequence of conditional branches, the code uses a state-machine-like structure. A state value is transformed through custom operations, including byte swapping, rotations, inversion, and negation. The result selects a location through a jump table.
This makes the code's execution flow harder to reconstruct using conventional static analysis.
Defensive focus: Use dynamic analysis, execution tracing, and memory inspection when a suspicious binary's logic is obscured by indirect control transfers.
4.4 Modified Meterpreter Payload
The decrypted payload is a backdoor with structural similarities to stageless shellcode associated with the Metasploit Framework.
Sophos identified similarities involving position-independent execution, PE-structure validation, and in-memory loading of executable content.
However, the payload differed from standard Metasploit shellcode in its implementation. Researchers identified non-standard API hashing and transformation routines rather than the familiar ROR13 method associated with Metasploit API resolution.
The analyzed payload was also larger and more complex than typical Metasploit shellcode. Sophos identified versions 1.1.5 and 1.1.7.
These findings indicate that the payload should not simply be classified as an unmodified, standard Meterpreter sample.
4.5 COFFLoader-Related Functionality
The analyzed samples contained elements associated with COFFLoader.
Sophos researchers interpreted the presence of standard APIs associated with the project as suggesting that the malware author may have used a custom Beacon Object File (BOF) to implement additional post-exploitation functionality.
The research also identified in-memory strings associated with C++ function decorators that demonstrated data-stealing capabilities built into the payload.
This suggests that the backdoor's purpose may extend beyond establishing access to the compromised system.
4.6 Host Profiling and Data Collection
The backdoor sends a structured JSON object containing a nested info block to its C2 server.
According to Sophos, the collected information can include:
Hostname and username.
GUID and domain information.
Operating-system version.
Internal and public IP addresses.
Process listings and metadata.
Process IDs and process names.
Working directory.
Code page.
CPU architecture.
Implant start time.
Implant type, such as server or client.
This information can help an operator understand the compromised host and identify additional targets.
Defensive focus: Investigate unexpected processes that collect extensive host information and send structured data to external destinations. Correlate the network activity with the process that generated it.
4.7 Command-and-Control Communications
Sophos observed payloads contacting the following legitimate services:
file[.]iogofile[.]ioably[.]netably[.]io
Researchers assessed that these connections may have been used to receive C2 tasking.
The research also identified two domains contacted by payload version 1.1.5:
dc[.]cathaypacifice[.]comnet[.]dpworldto[.]com
Payload version 1.1.7 was observed communicating with GitHub repositories to retrieve encrypted configuration files.
Connections to legitimate platforms do not automatically indicate malicious activity. Investigators should correlate destinations with the initiating process, request path, timing, and surrounding endpoint behavior.
4.8 GitHub-Based Configuration Retrieval
Sophos identified the following GitHub API repository path:
hxxps://api[.]github[.]com/repos/LatonyCharnleyaW
The repository contained files assessed by researchers as likely encrypted configuration files:
main.3ku8inqb.js.mapmain.lhxjkkai.js.mapmain.y8hnze7q.js.mapREADME.md
The repository also contained encrypted files with a .ts extension that may have represented implants used by the backdoor.
The use of a legitimate code-hosting platform to retrieve encrypted content can complicate network-based detection.
4.9 Network and Account Enumeration
Sophos observed the use of Windows net commands to enumerate users and network resources.
Such commands can help attackers identify accounts, shared resources, and systems of interest.
Defensive focus: Monitor unexpected enumeration commands, especially when executed by unfamiliar processes or users, or shortly after suspicious DLL loading.
4.10 Active Directory Database Access
Sophos observed the operators creating Volume Shadow Copies using vssadmin.exe before copying the ntds.dit file.
ntds.dit is the Active Directory database file. It contains directory information, including credential-related material and details about users, groups, and domain configuration.
Unauthorized access to this file can create serious identity-security risks. Depending on the access obtained and surrounding controls, attackers may be able to use extracted information to support further compromise.
Defensive focus: Investigate unexpected vssadmin.exe activity, suspicious shadow-copy creation, and access to ntds.dit. Review the initiating account, process tree, command-line arguments, and subsequent file operations.
5. Victimology and Attribution
5.1 Geographic Targeting
Sophos analyzed incidents occurring between August 2025 and August 2026 and identified victims in:
The Philippines.
Taiwan.
Afghanistan.
Namibia.
Likely South Korea.
The activity spanned organizations in Asia and Africa.
5.2 Attribution Assessment
Sophos assessed that the targeting patterns were consistent with Chinese intelligence-gathering interests. Researchers also observed activity concentrated during UTC+8 hours.
The geographic targeting overlapped with activity associated with other China-linked groups, including BRONZE PRESIDENT and BRONZE ATLAS. However, Sophos could not reliably link the investigated activity to either group.
BRONZE SANDCASTLE was therefore maintained as a separate designation.
These findings should be described as Sophos's assessment. Geographic targeting and operating-hour patterns alone do not conclusively establish the identity or nationality of an attacker.
6. Indicators of Compromise (IOCs)
The following indicators are drawn from the SophosLabs IOC CSV supplied for this article. The hashes and descriptions below are retained from that file. <FileCite ref_id="turn1file0" />
Important: Validate indicators against current telemetry and the original source before blocking or taking action. Filenames alone are not proof of compromise, and a hash identifies a particular file sample rather than every possible variant.
6.1 Command-and-Control Domains
| Domain | Source description |
|---|---|
dc[.]cathaypacifice[.]com | BRONZE SANDCASTLE C2 server |
net[.]dpworldto[.]com | BRONZE SANDCASTLE C2 server |
6.2 SHA-256 File Hashes
| SHA-256 | Associated artifact |
|---|---|
7afd292994ff214ca4dd880c834532274c957963e9838c7fa2e4d1318af75756 | sbiedll.dll loader |
b39e41be217fe2c690f75e9861cf64df52b302c66619fad608ca60a959288842 | jli.dll loader |
7515417e523a776e371877e868536fa1d1961826d4bc478734be993985d9df03 | Meterpreter payload in config.ini |
ab200eda30dd1868f597ac42e7876894c5838df84605bbd4fca8dbc91c90ea28 | libcef.dll loader |
dd9223c381893f493ca581e5bd35f16f778689e58fc459c1b0c39b0d33b41290 | CoreFoundation.dll loader |
95c6e9447585670b1a50551c36abe22154f6e010bfb23275fc467d2b20a423d9 | sbiedll.dll loader |
548d5add2b27ab8c2d506b81623a37f3d27a0f4e22e6fa938c6a008b70775989 | jli.dll loader |
592a6e74366bcd0c8a47657a88a6e81f99d2876baa6f923bb928bba4477054fc | NvdaHelperRemote.dll loader |
2332d924b1404961bc65280e782c787a4b048da7519fe73ccf4dc687b843d060 | jli.dll loader |
425177b18cb386f359796576e51666b7350489579550e564f51f8973b2f0979b | jli.dll loader |
72259376f63116d1aca64bf2ef42a5d211c1ebdb01516ef3b53778ed1347c238 | jli.dll loader |
2411a678e4de91451e69c27369bab4329c3cd969d4ebd851ba91dcfdba7e7db2 | MSIMG32.dll loader |
fdef7b1f4c013d686bf5dbfc92f37191c04beb0dc8b53c608bb8019a8c678052 | jli.dll loader |
e50dc9db568c272d7b019ec8e5fb1f111dd071baa4b3a00bc3c8c885ade46fbc | snmpapi.dll loader |
6c5c332428d914fd8eac0164947a70ec3d71bcd3f6cdcccb5e875dcb68ea076d | sbiedll.dll loader |
91ef27b1e83f324348ca51466ab17e2464f927b9c84758898fd7ce7ae6a6008b | sbiedll.dll loader |
437ba4b0db084c261f7843f85329b5204a2dedecdacf29b25a53a90bcf8d7d2e | jli.dll loader |
c7f1c6cfd87e4c4f492a6da634ae6a264467aaee5a3eed14e89c2297a58e5d30 | jli.dll loader |
c39ac4e1fe11639a9d46e026850cc3146eb809b542daf38e1a00eab11eaa1834 | sbiedll.dll loader |
e404b2b9f9f203dad81bdc1ef8bfcbe4f42eb616f113a89ed504202212af1c1e | jli.dll loader |
5e2b035ae40f8eaab40792a92573171df0011c2fbed55582268a824effa7cb91 | msvcp140.dll loader |
6.3 SHA-1 File Hashes
| SHA-1 | Associated artifact |
|---|---|
1f74838581e8bef12cb7df2d70099cd6865ef238 | sbiedll.dll loader |
e98c54ce736a607803e89144daa679fccd4f7521 | jli.dll loader |
f7db825aeee21d6326fa9c31512d090720546341 | Meterpreter payload in config.ini |
dae5a4489671e2d9d2189a830b17ff144f8ab904 | libcef.dll loader |
ffe58ea274af642a3fc1d50e823cb01f61dc8dae | CoreFoundation.dll loader |
cd1a262f8c2624700440b0aa464b7a2b42e23511 | sbiedll.dll loader |
536d6a8fa98a566079a459b16df1e29622dd8af5 | jli.dll loader |
4e6d600ce7a9cf6414e738d8f39260ae961602f0 | NvdaHelperRemote.dll loader |
624da53bd827ca27b66d8a42a9524b0b48e882b4 | jli.dll loader |
56b1258c4553fb549fca9215727be1e5cd1efe4e | jli.dll loader |
1abe25d3255679bdc1c35c070da40e3c4572bb4a | jli.dll loader |
2f399541445cc1aade0e7a2148853fe41be57827 | MSIMG32.dll loader |
120d204d253293d594d6fcbe681797942fb74f7a | jli.dll loader |
0ed62f00b9e7c5780c9bd05b096eee13a160480f | snmpapi.dll loader |
c2811a0fa539e4cd6633f32349fb34b0455bcba6 | sbiedll.dll loader |
1fc7350e215759b4d66ab6458244271b5e2cfa90 | sbiedll.dll loader |
24ede364f5eabe991c866e92ffa396bcb894eb14 | jli.dll loader |
6204c6342c5dfb67bdc29d45a524c3ac3a99aa01 | jli.dll loader |
ffd5cfbd5274a351315ba1e922d795f872f7ff9b | sbiedll.dll loader |
8964de054bfb70c5560ebabe224bb45723f60b37 | jli.dll loader |
ab86a6b71582e16a7da14533fe9407d7a2bc525c | msvcp140.dll loader |
6.4 MD5 File Hashes
| MD5 | Associated artifact |
|---|---|
07ed4b9470b2395ad5f969a501761ab8 | sbiedll.dll loader |
25ee87bb42d88bf00f60c4064c376846 | jli.dll loader |
dd89cfd1f0b13addfc43cae758130b6a | Meterpreter payload in config.ini |
615a6089ab073115ed0b8bcfcfbdc966 | libcef.dll loader |
4e916e024ea0c0eb16bd91f054d22444 | CoreFoundation.dll loader |
e63ec733393e4e7e3424cbff918003a3 | sbiedll.dll loader |
2a39552543f84aa3a7a562d9bb54fc00 | jli.dll loader |
8c32bf3711b237e3c781a00b5c526695 | NvdaHelperRemote.dll loader |
f771159881bc82217d30a64a7716e613 | jli.dll loader |
3addd7c803b9570fc0b1ee6899444c66 | jli.dll loader |
a0fa0893e1adde6b93063ab2c7369536 | jli.dll loader |
0a7464945e1994a7fc5ee2b86ce48c94 | MSIMG32.dll loader |
e43a7c1169cf1d4ce8b22c1301ca100c | jli.dll loader |
5ac863a8bffbcca5647284e088d401dc | snmpapi.dll loader |
b695542718a7cc373112852d97d1811e | sbiedll.dll loader |
4cf187200a4e639e314b7f96a33f161e | sbiedll.dll loader |
32151547d912488ac5242731057170a3 | jli.dll loader |
c15c10589e671094185311e358b8903b | jli.dll loader |
5910e874fccc9b37239b00b993393994 | sbiedll.dll loader |
0ff9ecbaf18e45cd7b4943b21d5c6cf1 | jli.dll loader |
ebf221383c234ea2361ed54e6df6ddfc | msvcp140.dll loader |
6.5 Sophos Countermeasures
Sophos lists the following detection names in its article:
ATK/Swrort-YATK/ScLoad-WATK/ScLoad-XATK/ScLoad-YATK/ScLoad-ZTroj/Loader-JLTroj/Loader-JP
These are Sophos-specific detection names. Other security products may use different labels or detection logic.
6.6 IOC Investigation Guidance
Security teams should use these indicators as starting points for threat hunting:
Search endpoint telemetry for the listed file hashes.
Investigate the associated DLLs and the executable processes that loaded them.
Review DNS, proxy, firewall, and endpoint logs for connections to the listed C2 domains.
Examine unexpected access to encrypted configuration files and suspicious in-memory execution.
Correlate IOC matches with account enumeration, process discovery, and Active Directory access.
Preserve suspicious files, logs, memory artifacts, and timestamps for forensic analysis.
Validate any match against the complete environment and relevant source data before blocking or isolating systems.
Full IOC source: SophosLabs — BRONZE SANDCASTLE Meterpreter IOCs
7. Potential Impact
7.1 Unauthorized System Access
The backdoor can provide operators with a way to interact with compromised systems and perform additional actions. The resulting impact depends on the privileges obtained and the systems accessible from the initial host.
7.2 Exposure of Host and Network Information
The payload collects operating-system details, network addresses, process metadata, and other information that can help attackers understand the victim's environment.
7.3 Active Directory Compromise
Access to ntds.dit presents a serious identity-security risk. Unauthorized access to the Active Directory database can expose sensitive directory information and potentially support additional credential-related attacks.
7.4 Data Theft
Sophos identified data-stealing capabilities within the analyzed payload. Organizations should determine whether sensitive files, identity data, or other information were accessed and transferred outside their environment.
7.5 Reduced Detection Opportunities
Encrypted payloads, control-flow obfuscation, and in-memory execution can reduce the effectiveness of security tools that rely primarily on static file signatures.
7.6 Operational and Reputational Consequences
A successful intrusion may lead to incident-response costs, business disruption, sensitive-data exposure, regulatory scrutiny, and reputational harm. The exact consequences depend on the scope of the compromise.
8. Recommendations
8.1 Immediate Actions
1. Investigate suspicious DLL loading
Review endpoint telemetry for trusted applications loading DLLs from unexpected directories. Check file paths, signatures, hashes, parent processes, and execution context.
2. Hunt for encrypted configuration files
Search for suspicious config.ini files and archives containing unexpected executable/DLL/configuration combinations. Correlate file access with subsequent process and network activity.
3. Monitor in-memory execution
Use EDR and memory-analysis capabilities to identify suspicious reflective loading, anomalous PE structures in memory, and processes that decode files before executing code.
4. Review outbound communications
Investigate connections to the C2 domains and other services described by Sophos. Attribute connections to their originating processes and investigate unusual repository paths or download activity.
5. Examine reconnaissance commands
Review command-line logs for unexpected Windows net commands, account enumeration, network discovery, and suspicious process listings.
6. Protect Active Directory
Investigate unexpected use of vssadmin.exe, Volume Shadow Copy creation, and access to ntds.dit. Verify that domain controllers and their backups are protected by strict access controls.
7. Preserve evidence
Retain suspicious DLLs, encrypted configuration files, endpoint logs, process trees, relevant memory artifacts, and network records.
8. Contain confirmed compromises
Isolate affected systems as appropriate, restrict compromised accounts and access paths, and determine whether the activity has spread to other endpoints or identity infrastructure.
8.2 Preventive Actions
A. Endpoint and Behavioral Monitoring
Enable telemetry for process creation, command-line execution, and DLL loading.
Monitor unexpected DLLs loaded from writable or untrusted locations.
Investigate suspicious in-memory execution and reflective loading.
Use behavioral analytics alongside file-based detection.
Correlate process, identity, file-system, and network events.
B. Application and DLL Security
Restrict execution of unapproved software and libraries where feasible.
Apply application-control policies and allowlisting where operationally appropriate.
Monitor trusted applications loading unexpected DLLs.
Maintain an inventory of approved software and expected library dependencies.
Review software deployment practices to reduce unauthorized executable/DLL combinations.
C. Identity and Active Directory Security
Apply least privilege to domain and server administration.
Restrict access to domain controllers and sensitive directory files.
Monitor unexpected access to
ntds.ditand other credential stores.Audit privileged accounts and suspicious domain-configuration changes.
Enforce MFA for remote and privileged access wherever supported.
Maintain secure, tested backups of critical identity infrastructure.
D. Network Security
Monitor unusual outbound communications and unexpected use of file-sharing, messaging, and code-hosting platforms.
Correlate DNS, proxy, firewall, and endpoint logs.
Restrict outbound traffic according to business requirements where feasible.
Investigate connections made by unfamiliar processes after suspicious DLL loading.
Maintain an incident-response process for suspicious C2 activity.
E. Incident Response Readiness
Maintain procedures for backdoor deployment, identity compromise, and data theft.
Establish evidence-preservation and memory-acquisition procedures.
Define containment and recovery priorities for domain controllers and other critical systems.
Conduct threat-hunting exercises focused on DLL sideloading and in-memory execution.
Update detections as new indicators and behavioral findings become available.
9. Conclusion
BRONZE SANDCASTLE demonstrates how DLL sideloading, encrypted configuration files, runtime decoding, reflective loading, and modified Meterpreter-based backdoors can be combined to support cyber-espionage operations.
The use of trusted executables paired with malicious DLLs is a distinctive aspect of the activity. Encrypted payloads and in-memory execution complicate static detection, while host profiling, network reconnaissance, and access to Active Directory data can expose sensitive information and increase the risk of wider compromise.
Sophos observed victims across Asia and Africa and assessed that the targeting was consistent with Chinese intelligence-gathering interests. However, the researchers did not reliably link the investigated activity to other similarly tracked groups. Technical evidence and attribution assessments should therefore remain clearly distinguished.
Organizations should monitor DLL loading and memory execution, investigate suspicious external communications, protect Active Directory, and correlate endpoint and network telemetry. Combining behavioral detection with least-privilege access and strong incident-response procedures provides a more resilient defense than relying on file signatures alone.
