Description
The incident highlights the risk of sensitive government data being copied from systems associated with public services and subsequently linked to infrastructure operated by an attacker.
Hunt.io discovered an exposed working directory containing the operator's own transfer records and supporting tools. The logs documented the completed transfer of 27 SQL Server backups, with filenames referencing systems such as POEA eServices, helpdesk services, overseas Filipino worker (OFW) databases, accreditation, repatriation, and electronic contracts.
These filenames suggest a connection to the Philippines' overseas-employment administration, whose functions were integrated into the Department of Migrant Workers (DMW). Hunt.io also linked two addresses that retrieved the operator's tools to DMW-related portal domains.
However, the researchers did not inspect the backup contents, and the recovered evidence did not establish the exact number of individuals whose information may be affected. The incident should therefore be described as a reported theft of database backups, not as a confirmed disclosure of a specific number of personal records.
Technical Description
The investigation identified three principal components in the exposed staging environment.
1. Rebranded GodPotato payload
The operator staged two .NET builds of a modified version of GodPotato under the filenames Wrqsnx.dll and Wrqsnx_portal.dll. GodPotato is a publicly available Windows privilege-escalation tool that can abuse certain service-account privileges to obtain SYSTEM-level access.
The recovered versions had renamed namespaces, classes, named-pipe paths, and assembly metadata. Hunt.io found evidence that the payloads had been rebuilt using different compilers.
The payload's presence establishes that the tool was staged, but does not prove it executed on a victim system.
2. PowerShell stager
The file portalload.ps1 was designed to collect Base64-encoded payload fragments, reconstruct them in memory, load the resulting .NET assembly, and invoke its entry point.
This approach avoids writing the reconstructed DLL directly to disk. It does not eliminate all forensic evidence: PowerShell activity, process creation, command-line arguments, and related telemetry may still be available to defenders.
Hunt.io recovered the stager but did not confirm its execution on the affected systems.
3. Impacket toolkit
The exposed directory contained an Impacket environment with scripts and utilities associated with Windows network protocols, Active Directory operations, credential access, and remote execution.
Researchers also identified scripts for checking LDAP and Microsoft SQL Server relay conditions and for filtering domain controllers by operating-system version.
The toolkit indicates the operator had access to capabilities relevant to Windows network environments. Hunt.io did not find evidence proving that these utilities were executed against the agency.
Attack Overview
1. Database backup collection
On September 6, 2026, the operator's transfer log recorded the copying of 27 SQL Server backup files totaling 68.837 GiB. The logged transfer completed in 23 minutes and 46 seconds.
The filenames referred to overseas-employment and migrant-worker systems. The available evidence did not reveal the exact source host or the destination where the files were stored.
2. Exposure of the operator's staging server
The operator hosted a working directory on an UpCloud server in Singapore at 213.163.205[.]35. A Python HTTP server exposed the directory over port 80 without authentication.
The directory included transfer logs, malware-related files, source code, a PowerShell stager, and the Impacket environment. Hunt.io captured the material while it remained accessible.
3. Recovery of tools and payloads
The access logs showed that the stager, two payloads, and source archive were retrieved later on September 6. The first payload was fetched approximately 20 hours after the recorded database transfer had completed.
This sequence is important: the recovered logs do not establish that the staged payloads were responsible for the database theft. They indicate that the operator already had sufficient access to copy the backups before those payloads were retrieved from the staging server.
4. Infrastructure correlation
Hunt.io correlated two addresses that retrieved the tools with the domains apps.dmw.gov[.]ph and legalassistance.dmw.gov[.]ph. Certificate information also linked the associated infrastructure to the Philippine Overseas Employment Administration.
These correlations strengthen the assessment that the activity involved infrastructure associated with overseas-employment services. They do not independently establish the full scope of compromise or identify the person operating the systems.
5. Investigation and attribution
Hunt.io reported notifying the relevant national computer emergency response team before publication.
The researchers did not identify the initial access vector, establish the ultimate destination of the copied backups, or confirm execution of the recovered payloads. They also explicitly declined to attribute the incident to a known threat actor or country.
Technical Analysis
Completed Data Transfer
The strongest evidence in the investigation is the operator's own transfer log. It records 27 files and a completed transfer of 68.837 GiB.
This provides direct evidence of the copying operation as recorded by the tool. However, the log does not identify the source system or final destination, and the backup contents were not inspected. It therefore cannot independently establish the exact personal information involved or how that information was subsequently used.
Open Directory and Operational Exposure
The staging server's directory listing exposed operational records and tooling to unauthenticated visitors. Hunt.io recorded 543 requests from 76 distinct IP addresses over approximately 16 hours.
This illustrates how an operator's own infrastructure can become a source of evidence. Transfer logs, filenames, payloads, and source archives can reveal activity that might otherwise be difficult to reconstruct.
Not every visitor to the directory was involved in the operation. Hunt.io distinguished the operator's staging server from addresses that retrieved tools and other systems that appeared to be ordinary scanners.
Privilege Escalation Through GodPotato
GodPotato can exploit a Windows service account's existing privileges, particularly the ability to impersonate security tokens, to obtain SYSTEM-level execution under suitable conditions.
The recovered payload used RPC and named-pipe interactions to obtain access to a SYSTEM token. Hunt.io noted that the technique did not require a Windows software vulnerability in the described scenario; it depended on available privileges and the host's security context.
The rebranded payload demonstrates preparation for possible privilege escalation. It does not prove successful escalation on a government system because the report contained no confirmed execution telemetry.
In-Memory Payload Loading
The PowerShell stager reconstructed a .NET payload from Base64 fragments and loaded it into memory. This technique can reduce the presence of standalone executable files on disk, but should not be considered invisible.
Defenders can investigate suspicious PowerShell execution, access to payload fragments, unusual .NET assembly loading, unexpected command execution, and abnormal interactions with Windows RPC and named pipes.
Attribution Limitations
The recovered source files contained Chinese-language comments, and the archive metadata included the account name cicada. Hunt.io cautioned that these clues were insufficient to establish an operator's location, identity, or affiliation.
Language in a development environment can reflect a toolchain or third-party source code rather than the person using it. No state nexus or known threat actor was confirmed.
Indicators of Compromise (IOCs)
Hunt.io published network indicators, file hashes, filenames, and host artifacts. These should be interpreted according to their documented roles: some identify the exposed staging infrastructure, while others identify recovered tools. Several IP addresses had ambiguous roles and must not automatically be treated as attacker infrastructure.
Network indicators
| Indicator | Reported context |
|---|---|
213.163.205[.]35 | Exposed staging server hosting the unauthenticated directory on TCP port 80 |
172.105.82[.]111 | Retrieved tools from the directory; role not established |
156.59.127[.]203 | Retrieved the source archive and a payload; role not established |
121.58.231[.]x | Masked network range associated with requests for the loader, payloads, and source archive; exact address withheld |
158.178.237[.]116 | Swept the directory and retrieved files; assessed as scanner activity rather than confirmed operator activity |
Associated services reported by Hunt.io:
TCP port 80 — unauthenticated Python HTTP directory server.
TCP port 8888 — Python upload receiver that wrote request bodies to
/tmp/uploaded.dll.
SHA-256 file hashes
| File | SHA-256 |
|---|---|
Wrqsnx.dll | 4188295ae1f7f8a99d15e90eb037eef3eb03f6acd4422bf877c6e3304e70fcb9 |
Wrqsnx_portal.dll | 0405d4f79a3cc97fa492f1957cde0d51bc569f5432792f1b8f72077aa3fb2434 |
portalload.ps1 | fcbfc74ddffad4789120c618bdf166c6c8aaca286fb2d184f7fa0ed8b8f26c59 |
upload.py | cde477ef57ef6f100dcb0242f055113690fdeade8204e0986a3ad257be1fb81e |
gp_src.tgz | 9d3f2f6031ec3b487ae12dd6f9311dd89b4a6bd23af252212bcc1f5507784f39 |
http80.log | ee15c2a54f1bb483f70bf7777ae08d39c8abbbf7f5b1f2139de0ff76ec3e241e |
rclone.log | e7e6bfc3a2d6c3eb0ed6e1f4ade99ca761d517886412da99a2cf4a0626bce5e3 |
Host artifacts and strings
C:\Users\Public\ptcmd.txt— command input referenced by the payload; not confirmed on a victim host.C:\Users\Public\p_*.b64— Base64 payload fragments used by the stager.\.\pipe\Wrqsnx\pipe\epmapper— named-pipe path associated with the rebranded payload.POTATO_FIRED— stager completion marker.[+] SHELL_DONE— payload output marker.
The presence of these artifacts in the recovered staging directory does not establish that they were present on, or executed by, an affected government system.
Potential Impact
If the reported theft involved operational government databases, the incident could have implications beyond the immediate loss of data.
Exposure of sensitive records: Database backups associated with overseas-employment systems could contain personal, employment, contractual, or administrative information. Their exact contents remain unverified.
Privacy and identity risks: If personal records were present and subsequently misused, affected individuals could face targeted phishing, impersonation, or identity-related fraud.
Operational confidentiality: Information about accreditation, repatriation, case management, or contracts could reveal sensitive administrative processes.
Follow-on intrusion risk: The recovered toolkit could support privilege escalation and Windows network operations if successfully deployed in an environment.
Compliance and trust: A confirmed exposure of personal data could create regulatory obligations and undermine confidence in government services.
The report does not establish the number of affected individuals, confirm the contents of the backups, or demonstrate subsequent misuse of the information.
Recommendations
Immediate Actions
Preserve relevant logs: Secure SQL Server, application, authentication, firewall, endpoint, and file-transfer logs. Prioritize evidence covering September 6–7, 2026, while retaining a wider investigation window where available.
Validate backup access: Determine which systems stored or served the affected backup files, review access records, and establish whether unauthorized copies were created.
Investigate suspicious infrastructure access: Examine outbound transfers, unusual service activity, and connections involving the reported staging infrastructure. Treat third-party addresses according to their documented roles rather than blocking every address indiscriminately.
Hunt for the recovered artifacts: Search endpoint and server telemetry for the published SHA-256 hashes,
Wrqsnxfilenames,portalload.ps1, related payload fragments, and named-pipe artifacts.Review PowerShell and token activity: Investigate suspicious PowerShell processes, in-memory .NET loading, unusual RPC interactions, token impersonation, and unexpected SYSTEM-level command execution.
Assess service-account privileges: Identify Windows service accounts with impersonation privileges and confirm that their rights are appropriate for their roles.
Rotate exposed credentials where warranted: If the investigation finds that credentials were accessible or compromised, rotate them in accordance with incident-response procedures and assess dependent services.
Coordinate disclosure and response: Work with the relevant national CERT, privacy authorities, affected agencies, and legal teams to establish the facts and determine any required notifications.
Preventive Measures
Secure database backups: Restrict backup access to authorized systems and identities, encrypt backups at rest and in transit, and monitor bulk reads and transfers.
Harden staging infrastructure: Disable unauthenticated directory listings, restrict HTTP access, and avoid storing operational logs, payloads, or sensitive artifacts on publicly accessible servers.
Apply least privilege: Minimize service-account permissions, particularly impersonation rights, and review configurations that could enable token abuse.
Improve endpoint detection: Monitor for suspicious PowerShell activity, reflective .NET loading, unusual named pipes, and abnormal process execution under SYSTEM.
Monitor data movement: Alert on large database exports, unusual outbound transfers, and unexpected access to backup repositories.
Segment critical systems: Restrict communication between public-facing services, database environments, administrative servers, and identity infrastructure.
Maintain tested recovery plans: Regularly verify backup integrity, access controls, restoration procedures, and incident-response playbooks.
Conclusion
The Hunt.io investigation documents a completed transfer of 68.837 GiB of database backups associated by filename with Philippine overseas-employment and migrant-worker systems. The discovery of the operator's exposed staging directory provided investigators with transfer logs, tooling, and file-level indicators that helped reconstruct part of the activity.
However, important questions remain unanswered. The initial access method, final destination of the backups, exact contents of the data, and execution of the recovered payloads were not established. Hunt.io also did not attribute the operation to a known threat actor or state.
The incident demonstrates the value of protecting backup repositories, monitoring bulk data movement, limiting service-account privileges, and treating exposed operational infrastructure as a potential source of forensic evidence.
