1. Introduction
Swallowtail, also known as APT28, Fancy Bear, and Forest Blizzard, is a long-running Russian cyberespionage group associated with Russia's military intelligence agency, the GRU. The group primarily targets military, defense, and government organizations across Europe and North America to obtain intelligence relevant to national security.
On October 20, 2026, a report by Symantec's Threat Hunter Team described Hookedge backdoor activity targeting organizations in two NATO member states. The victims included a law enforcement agency, a naval shipbuilder, and a manufacturer of military aircraft engines.
The intrusion against the naval shipbuilder involved repeated operator activity over at least two months. The attackers stole credential material, established a persistent SSH tunnel into the internal network, conducted reconnaissance, and deployed a previously undocumented reflective loader named ld.exe.
The campaign demonstrates how targeted espionage operations can combine malicious documents, legitimate system utilities, custom malware, credential theft, and covert remote access to maintain a foothold in sensitive environments.
2. Key Highlights
Threat actor: Swallowtail, also known as APT28, Fancy Bear, and Forest Blizzard.
Attribution: Associated with Russian military intelligence, according to the cited reporting.
Targeted sectors: Law enforcement, naval shipbuilding, aerospace, military engine manufacturing, government, and defense.
Primary malware: Hookedge backdoor.
Initial delivery: A malicious, macro-enabled Microsoft Word document received as an email attachment.
Credential theft: An SSH private key was planted on a compromised host, while a PowerShell script captured NTLM authentication material.
Persistent access: An SSH tunnel with remote port forwarding provided a route from attacker-controlled infrastructure into the victim network.
Undocumented tool:
ld.exe, a reflective loader that decrypts an encrypted file in memory.Command-and-control infrastructure: The campaign used webhook.site and the IP address
185.82.218[.]119.Payload status: The final encrypted payload was not recovered, so its ultimate functionality remains unknown.
3. Threat Actor Profile
Swallowtail is tracked under several names, including APT28, Fancy Bear, and Forest Blizzard. Public reporting associates the group with Russia's GRU military intelligence service.
The group has a history of targeting government, military, and defense-related organizations. It gained widespread public attention following its reported involvement in the 2016 Democratic National Committee compromise.
In this campaign, the victims shared a clear national-security connection. A law enforcement agency, a naval shipbuilder, and a military aircraft engine manufacturer could all possess information of intelligence value to a foreign government.
The targeting across multiple organizations and defense-related sectors suggests a broader intelligence-collection effort rather than an attack limited to a single organization.
4. Targeting and Campaign Overview
Symantec's Threat Hunter Team identified Hookedge-related activity against organizations in two NATO member states. The observed targets included:
A law enforcement agency.
A naval shipbuilding organization.
A manufacturer of military aircraft engines.
The Hookedge backdoor had previously been documented by Recorded Future's Insikt Group, which tracks the activity as BlueDelta, in August 2026.
The intrusion at the naval shipbuilder provides a detailed view of the attackers' post-compromise operations. Following the initial infection, operators repeatedly returned to the compromised system, issued commands, collected information, stole credentials, and established a network tunnel.
The attackers later deployed ld.exe with an encrypted payload. Although the loader's behavior was analyzed, the encrypted payload itself was not recovered, limiting conclusions about its final purpose.
5. Attack Chain
Stage 1: Initial Access Through a Malicious Document
On July 24, 2026, a malicious Microsoft Word document was opened directly from the Outlook attachment cache. This indicates that the document had arrived as an email attachment.
The document contained a heavily obfuscated VBA macro. Rather than storing all of its code and strings in straightforward form, the macro split its contents into short fragments and reconstructed them at runtime, making static analysis more difficult.
After execution, the macro's AutoOpen routine dropped four files into the user's profile:
A VBScript launcher.
An HTML file.
An XHTML file.
A batch script.
Together, these files initiated the Hookedge infection chain.
Stage 2: Hidden Browser-Based Payload Retrieval
The batch script terminated running Microsoft Edge processes and removed stale download artifacts. It then launched Edge in a nearly invisible, one-pixel window positioned off-screen.
The browser received an inline, Base64-encoded JavaScript payload through a data:text/html;base64 URL. The JavaScript fetched a remote resource from webhook.site and saved the response to disk.
This technique used a legitimate browser as an HTTP client instead of relying exclusively on a conventional downloader.
The script subsequently located the downloaded file, copied its contents into a new command script, executed it, uploaded the output to the webhook endpoint, and deleted temporary files.
This created a delivery-and-callback loop through which the operators could send tasks to the compromised host and receive the results.
Stage 3: Initial Reconnaissance
The first downloaded task was a reconnaissance batch script. It checked connectivity to several services and addresses, including Dropbox, Google, Microsoft's login portal, Mega, example.com, and a hard-coded IP address.
The script also examined:
Proxy and Web Proxy Auto-Discovery (WPAD) configuration.
Logged-on sessions.
IP configuration.
Operating-system and system information.
Running processes.
User profile directories.
Downloads, Desktop, and Documents directories.
Program Files directories.
These checks helped the operators understand the compromised host, its network environment, and potential outbound communication channels.
Stage 4: Transition to Dedicated Command-and-Control Infrastructure
After a gap of approximately three days, a new copy of the Hookedge VBScript launcher appeared on July 27, 2026, under a randomly generated filename.
The attackers then deployed a more capable batch script that fetched tasks directly from the dedicated IP address 185.82.218[.]119 over HTTPS on port 443.
The new tasking pattern involved retrieving commands, acknowledging receipt, executing the commands, uploading output, and removing temporary files.
Seven tasks were issued in rapid succession over approximately 30 seconds. This timing is consistent with interactive operator activity rather than a simple fixed-interval polling mechanism.
Stage 5: SSH Private-Key Theft and Persistent Tunneling
The attackers downloaded a file, renamed it id_rsa, and placed it in a directory under C:\ProgramData\Adobe\. They then modified its ownership and permissions.
The private key was used with the Windows OpenSSH client to connect to the attacker-controlled server. The connection disabled host-key checking and enabled legacy RSA algorithms.
The SSH command used remote port forwarding, the -N option to avoid starting an interactive shell, and the -f option to run the session in the background.
This established a persistent tunnel that provided a route from the attacker's infrastructure back into the compromised organization's internal network.
Subsequent DNS lookups and connectivity checks against internal hostnames suggest that the operators were mapping the internal environment and testing connectivity through or around the established tunnel.
On July 28, the operators terminated ssh.exe. The available evidence does not establish whether this was routine cleanup, tunnel rotation, or a decision that the tunnel was no longer required.
Stage 6: NTLM Credential Capture
The attackers also used a separate PowerShell-based technique to obtain authentication material.
The script created an HTTP listener on the compromised host's loopback address and attempted to trigger local authentication through a net use command. It was designed to capture NTLM authentication challenge-response material from the resulting exchanges.
Such material may support offline password cracking or, in suitable circumstances, authentication relay attacks. Whether either outcome was successfully achieved in this incident is not established by the available report.
This activity is important because credential exposure can enable further access even if the initial backdoor is later removed.
Stage 7: Further Internal Reconnaissance
Following the credential-capture activity, the operators performed additional discovery operations, including:
Account identity and privilege checks.
Group-membership queries.
DNS lookups of internal hostnames.
Connectivity tests to internal systems.
System configuration and process enumeration.
Installed Windows patch inventory.
The operators also queried information that could help identify servers and other important assets within the victim environment.
These actions are consistent with continued investigation of the compromised network and potential preparation for additional access.
Stage 8: Deployment of the Reflective Loader
On July 31, 2026, the operators staged two files on the compromised host: ld.exe and Events.dat.
The loader was executed with an encrypted file and a decryption key supplied as command-line arguments:
C:\ProgramData\ld.exe DE063CF3F2E2A83E7B1A277C052A340E C:\ProgramData\Events.dat
Analysis determined that ld.exe was a reflective loader. It decrypts the supplied file in memory and invokes an exported function named ExecuteLoader.
The decrypted content was reported to begin with the four-byte marker ADSA (0x41534441). However, the specific Events.dat payload was not recovered, and its final functionality remains unknown.
The use of an encrypted payload and in-memory loading can make investigation more difficult because the executed content may not appear as a conventional decrypted executable on disk.
Stage 9: Continued Presence and Later Activity
The final distinct operator command in the available telemetry was net user. The same loader hash and Hookedge launcher were still present on the compromised host in late August, and the launcher was observed again on September 24, 2026.
The available evidence does not determine whether the scheduled task continued to execute without receiving new commands or whether operators deliberately resumed activity.
Organizations investigating similar indicators should therefore examine the full timeline rather than assuming that the initial document execution represents the end of the intrusion.
6. Malware Analysis: Hookedge
Hookedge is a backdoor used to retrieve tasks and return command output. In the described campaign, it was delivered through a macro-enabled Word document and used browser functionality and scripting to establish its initial communication pattern.
Observed characteristics
Uses a VBScript launcher and supporting scripts.
Retrieves tasks through webhook.site during the initial stage.
Uses a hidden Microsoft Edge instance to fetch remote content.
Executes downloaded tasks through batch scripts.
Uploads command output to remote infrastructure.
Later communicates with a dedicated attacker-controlled IP address.
Supports repeated task execution and interactive operator activity.
The combination of scripting, browser processes, and remote tasking can complicate detection if defenders monitor only traditional executable malware.
7. Malware Analysis: ld.exe
The executable ld.exe is a previously undocumented reflective loader identified during the intrusion.
Observed characteristics
Accepts an encrypted file and a decryption key as arguments.
Decrypts the file in memory.
Invokes an exported function named
ExecuteLoader.Was executed with the encrypted file
Events.dat.Uses the SHA-256 hash listed in the IOC section below.
The encrypted payload was not recovered. Therefore, its final functionality, persistence behavior, and any additional malicious capabilities cannot be confirmed from the available evidence.
8. Indicators of Compromise (IOCs)
The following indicators were provided in the source report. They are defanged where appropriate for safe publication.
8.1 Network indicators
Attacker-controlled IP address
185.82.218[.]119
Webhook.site URLs
hxxps://webhook[.]site/890efd8d-6e9e-428a-bcf9-987745d71000hxxps://webhook[.]site/53379b62-8020-41cf-af57-effa6e1e8506hxxps://webhook[.]site/5b73dc69-d765-4440-a4d2-e9eb33283eb9hxxps://webhook[.]site/d828b9fc-5435-4b1c-881f-d6350c11d0b6hxxps://webhook[.]site/ca3c058e-3c88-449b-8299-822cd5d5e17c
8.2 Hookedge dropper file hashes
The following SHA-256 values were identified as Hookedge droppers:
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
8.3 Hookedge VBScript launcher file hashes
063930af6a4aad06d8f8aab43226bf6dcddd2c79aa505894ee95876cf9254543d74c65e6d35d3b2a80686f059590f919316ecaf4ed539192e904d675ba5c7a40aafbc096e20f259df9166f837d0baf7333f177e6f383ca7cb9215936ecbe7f5e
8.4 Reflective loader file hash
Filename: ld.exe
SHA-256:
f1cc91e7427a22027f65d31bbee19c54b602d08d7d93e4745b9cea59fa8132ff
8.5 Additional artifacts
Events.dat— encrypted payload supplied to the reflective loader; final contents not recovered.id_rsa— SSH private-key filename observed during the intrusion.
IOC handling note: Validate the indicators against the original report and internal telemetry before taking blocking or remediation action. The presence of an indicator should be investigated in context and does not independently prove compromise.
9. MITRE ATT&CK Mapping
The following techniques are suggested mappings based on the reported behaviors. Validate them against the official MITRE ATT&CK definitions before publication.
T1566.001 – Spearphishing Attachment: Malicious Word document delivered through email.
T1204.002 – Malicious File: User interaction with the attachment to initiate the infection chain.
T1059.003 – Windows Command Shell: Batch scripts used for task execution and system discovery.
T1059.001 – PowerShell: Script-based NTLM authentication capture.
T1105 – Ingress Tool Transfer: Downloading scripts, the SSH private key, and loader-related files.
T1021.004 – SSH: SSH connection and tunneling used for remote access.
T1557 – Adversary-in-the-Middle: Potential credential-relay relevance of captured NTLM material; the specific sub-technique should be validated.
T1620 – Reflective Code Loading: Loader decrypts and loads content in memory.
T1082 – System Information Discovery: Collection of system and operating-system details.
T1057 – Process Discovery: Enumeration of running processes.
T1016 – System Network Configuration Discovery: Inspection of IP, proxy, and WPAD configuration.
T1018 – Remote System Discovery: Discovery of internal systems through hostname lookups and connectivity checks.
These mappings describe observed or reasonably inferred behavior; they do not establish that every technique was successful.
10. Detection and Threat Hunting
Security teams should search for the following behavioral patterns.
Email and endpoint telemetry
Suspicious macro-enabled Word attachments.
Office documents launched from Outlook attachment-cache directories.
Unexpected execution of
wscript.exe,cscript.exe,cmd.exe, and PowerShell following document access.Microsoft Edge processes launched with
--window-size=1,1, off-screen coordinates, or embedded Base64 HTML content.Batch scripts that download remote content, execute it, upload output, and delete temporary files.
Network telemetry
Connections to
webhook.siteassociated with unusual script execution.Connections to
185.82.218[.]119, particularly from Windows endpoints.Outbound SSH connections from endpoints that do not normally require SSH access.
SSH sessions configured for remote forwarding or background execution.
Unusual HTTP or HTTPS traffic that repeatedly retrieves command files and uploads output.
Identity and credential telemetry
Unexpected creation or modification of SSH private-key files.
Private keys stored in unusual directories under
C:\ProgramData.Unusual local HTTP listeners combined with PowerShell and authentication-related commands.
NTLM authentication events that originate from unexpected processes or local endpoints.
Suspicious account, group-membership, and privilege enumeration.
File and persistence telemetry
Files named
ld.exe,Events.dat, and suspicious copies of VBScript launchers.Files matching the hashes in the IOC section.
Unexpected scheduled tasks or other persistence mechanisms associated with suspicious scripts.
Repeated execution of the same launcher across a prolonged timeline.
11. Mitigation and Incident Response
Isolate affected systems: If active compromise is suspected, contain the endpoint while preserving evidence needed for investigation.
Investigate the full timeline: Correlate email, endpoint, identity, DNS, proxy, firewall, and authentication records to determine initial access, subsequent execution, and possible lateral movement.
Block confirmed malicious infrastructure: Use the supplied IP and URL indicators in appropriate security controls after validation.
Remove malicious components: Identify and remediate Hookedge, suspicious scripts, the reflective loader, related payloads, and persistence mechanisms as part of a coordinated response.
Rotate exposed credentials: Reset potentially compromised passwords and replace affected SSH keys. Revoke unauthorized keys and review privileged account access.
Review SSH access: Investigate unexpected SSH services and sessions, especially those using remote port forwarding or unusual authentication options.
Investigate NTLM exposure: Determine whether captured authentication material could have been cracked or relayed. Apply suitable protections and restrict NTLM where operationally feasible.
Hunt for lateral movement: Expand the investigation to other endpoints, servers, administrative accounts, and defense-related systems accessible from the compromised host.
Review persistence: Examine scheduled tasks, startup locations, script execution history, and repeated launcher activity.
Preserve forensic evidence: Retain suspicious files, command-line telemetry, logs, and relevant network records for deeper analysis.
12. Why This Threat Matters
This campaign is significant because it combines targeted espionage with techniques that can maintain access beyond the initial infection.
The Hookedge backdoor provided remote tasking, while the SSH tunnel created a route into the victim's internal network. The separate NTLM capture technique introduced another potential path to credential compromise. Finally, the reflective loader provided a way to decrypt and execute an unknown payload in memory.
For defense manufacturers, shipbuilders, aerospace companies, and government agencies, such access could expose sensitive operational, engineering, personnel, or strategic information. The available reporting does not establish exactly what information was exfiltrated or what the final payload was intended to do, so those outcomes should not be presented as confirmed.
The extended timeline reinforces the need to investigate for persistence and continued access rather than relying solely on a one-time malware scan.
13. Conclusion
The Swallowtail campaign highlights the risks posed by persistent, state-linked cyberespionage against NATO defense and government organizations. Through malicious documents, the Hookedge backdoor, credential theft, SSH tunneling, reconnaissance, and a previously undocumented reflective loader, the attackers demonstrated a layered approach to intrusion and remote access.
Organizations should prioritize endpoint and network hunting, credential security, SSH configuration review, and investigation of suspicious script execution. Any confirmed indicator should trigger a broader assessment of the affected host and its connections to internal systems.
Because the final encrypted payload was not recovered, the ultimate capabilities and objectives of that component remain unknown. Effective defense therefore requires correlating the known indicators with behavioral evidence and investigating the full scope of the intrusion.
