The massive amount of junk code that hides the malware's logic from security scans was almost certainly generated by AI, researchers say.
Researchers have uncovered a new malware strain capable of stealing credentials immediately after gaining a foothold on a victim network, capturing both stored browser passwords and live keystrokes in real time through a standalone stealer and a malicious browser extension.
What makes the malware particularly difficult to contain, according to ReliaQuest, is its likely use of AI-generated code and process injection to evade detection tools. It also incorporates a persistence mechanism that can silently re-execute even after an infected host appears fully clean.
"DeepLoad steals credentials from the moment it lands, so even partial containment can still leave you with exposed passwords, session, and active accounts," ReliaQuest warned in a report this week. "Before the main attack chain finishes, a standalone credential stealer, filemanager.exe, is already running on its own infrastructure and can exfiltrate data even if the main loader is detected and blocked."
In addition, the browser extension that the malware drops and registers can capture credentials in real-time as users type them, and it persists across browser sessions until explicitly removed, the security vendor said.
As with most ClickFix scams, the attack chain begins with users receiving fake browser prompts asking them to execute a seemingly benign command to "fix" some kind of made up "error." In this instance, the command immediately creates a scheduled task to re-execute the loader, so it persists across reboots or partial detection, without the user having to do anything thereafter. The malware then uses mshta.exe, a legitimate Windows utility, to communicate with the attacker's infrastructure and download a heavily obfuscated PowerShell loader.
DeepLoad's actual attack logic itself is a short decryption routine that unpacks its malicious payload entirely in memory. Once unpacked, the payload is injected into LockAppHost.exe, a legitimate Windows process that manages the lock screen and is something that most security tools do not actively monitor.
To carry out the injection DeepLoad uses a PowerShell feature called Add-Type to generate a temporary Dynamic Link Library (DLL) that is dropped into the compromised computer's Temp directory. ReliaQuest found the malware compiling the DLL freshly on every execution and assigning a randomized filename to ensure that security tools scanning for a specific filename would never find anything to match. The malware also disables PowerShell command history to cover its own tracks.
In the campaign that ReliaQuest investigated, DeepLoad also spread to connected USB drives within 10 minutes of initial infection. The security vendor found the malware writing more than 40 files disguised as Chrome setup files, Firefox installers, AnyDesk shortcuts and other familiar installers, to the USB drive of a compromised host.
The goal likely was to maximize the chances of a user clicking on one of the spurious installers and infecting another machine in the process. Even so, it is unclear whether USB propagation is a built-in feature of DeepLoad or if it's something that the attacker layered on for that particular campaign, ReliaQuest said.
The company recommended that organizations hit by DeepLoad audit and remove WMI event subscriptions on affected hosts before returning them to production. They should also enable PowerShell Script Block Logging and behavioral endpoint monitoring to look for malicious activity since file-based scanning will not detect the loader. In addition, organizations should change all credentials associated with a compromised system, including saved passwords, active session tokens, and accounts that were in use during the infection period.
"The indications of AI generation mean there is a realistic probability that obfuscation will evolve from generic noise to padding tailored to the specific environment it's deployed in, making behavioral baselining harder over time," ReliaQuest warned. "As WMI subscriptions are added to remediation checklists, the persistence mechanism is likely to shift to other legitimate Windows features that currently receive less attention."
Researchers have uncovered a new malware strain capable of stealing credentials immediately after gaining a foothold on a victim network, capturing both stored browser passwords and live keystrokes in real time through a standalone stealer and a malicious browser extension.
What makes the malware particularly difficult to contain, according to ReliaQuest, is its likely use of AI-generated code and process injection to evade detection tools. It also incorporates a persistence mechanism that can silently re-execute even after an infected host appears fully clean.
DeepLoad Delivery via ClickFix
The authors of the malware, which ReliaQuest is tracking as "DeepLoad" are using the ClickFix social engineering technique to distribute the credential stealer in enterprise environments."DeepLoad steals credentials from the moment it lands, so even partial containment can still leave you with exposed passwords, session, and active accounts," ReliaQuest warned in a report this week. "Before the main attack chain finishes, a standalone credential stealer, filemanager.exe, is already running on its own infrastructure and can exfiltrate data even if the main loader is detected and blocked."
In addition, the browser extension that the malware drops and registers can capture credentials in real-time as users type them, and it persists across browser sessions until explicitly removed, the security vendor said.
As with most ClickFix scams, the attack chain begins with users receiving fake browser prompts asking them to execute a seemingly benign command to "fix" some kind of made up "error." In this instance, the command immediately creates a scheduled task to re-execute the loader, so it persists across reboots or partial detection, without the user having to do anything thereafter. The malware then uses mshta.exe, a legitimate Windows utility, to communicate with the attacker's infrastructure and download a heavily obfuscated PowerShell loader.
Heavily Padded Loader
ReliaQuest's analysis of DeepLoad showed its functional code is buried under thousands of lines of junk code that appeared designed to overwhelm static scanning tools and leave them with nothing to flag. The sheer volume of padding in the loader suggests that it was not written by a human author, but most likely developed by an AI model, the security vendor said.DeepLoad's actual attack logic itself is a short decryption routine that unpacks its malicious payload entirely in memory. Once unpacked, the payload is injected into LockAppHost.exe, a legitimate Windows process that manages the lock screen and is something that most security tools do not actively monitor.
To carry out the injection DeepLoad uses a PowerShell feature called Add-Type to generate a temporary Dynamic Link Library (DLL) that is dropped into the compromised computer's Temp directory. ReliaQuest found the malware compiling the DLL freshly on every execution and assigning a randomized filename to ensure that security tools scanning for a specific filename would never find anything to match. The malware also disables PowerShell command history to cover its own tracks.
In the campaign that ReliaQuest investigated, DeepLoad also spread to connected USB drives within 10 minutes of initial infection. The security vendor found the malware writing more than 40 files disguised as Chrome setup files, Firefox installers, AnyDesk shortcuts and other familiar installers, to the USB drive of a compromised host.
The goal likely was to maximize the chances of a user clicking on one of the spurious installers and infecting another machine in the process. Even so, it is unclear whether USB propagation is a built-in feature of DeepLoad or if it's something that the attacker layered on for that particular campaign, ReliaQuest said.
Standard Remediation Not Enough
Standard cleanup, like removing scheduled tasks, temporary files, and other indicators of compromise (IOCs) is not enough to fully remediate DeepLoad infections, ReliaQuest said. That's because DeepLoad creates a persistent trigger within Windows Management Instrumentation (WMI) that automatically reruns the attack without any further user interaction. In the incident that ReliQuest investigated, the malware re-executed the attacks a full three days after the affected host had appeared to be fully cleaned.The company recommended that organizations hit by DeepLoad audit and remove WMI event subscriptions on affected hosts before returning them to production. They should also enable PowerShell Script Block Logging and behavioral endpoint monitoring to look for malicious activity since file-based scanning will not detect the loader. In addition, organizations should change all credentials associated with a compromised system, including saved passwords, active session tokens, and accounts that were in use during the infection period.
"The indications of AI generation mean there is a realistic probability that obfuscation will evolve from generic noise to padding tailored to the specific environment it's deployed in, making behavioral baselining harder over time," ReliaQuest warned. "As WMI subscriptions are added to remediation checklists, the persistence mechanism is likely to shift to other legitimate Windows features that currently receive less attention."