This Malware Hides Its Code on Your GPU

TL;DR: An old USB worm is back, spreading new malware called CoffeeLoader. It cleverly uses a computer's graphics card (GPU) to decrypt its malicious code, a technique designed to evade standard security software and analysis tools.
Key facts
- Category
- Cybersecurity
- Impact
- High
- Published
- Source
- SecureList
Full summary
A resurgent USB worm is spreading new malware that uses your graphics card to hide its malicious activity from security tools.
Security researchers at SecureList have identified a new malware campaign that combines an old-school infection method with a highly advanced evasion technique. An older USB worm known as Retadup has been repurposed to distribute a new malware strain called CoffeeLoader. The worm spreads through removable drives, a vector that many organizations may no longer prioritize in their threat models. Once executed on a victim's machine, it begins a sophisticated process to install the final payload, demonstrating a significant evolution in how attackers are bypassing modern security defenses. The campaign appears to be a targeted effort, leveraging a once-common threat to deliver a payload designed to operate deep within a system, far from the eyes of conventional security software.
The most innovative aspect of CoffeeLoader is how it hides its malicious code. According to the analysis, the initial installer drops a modified version of a legitimate library file (`7z.dll`) associated with the 7-Zip archiving tool. This trojanized file contains shellcode that initiates the decryption of the main payload. Instead of performing this decryption on the computer's main processor (CPU), where security tools actively monitor for suspicious activity, CoffeeLoader offloads the entire process to the graphics processing unit (GPU). By using the GPU's computational power, the malware keeps its decrypted, malicious components out of the system's main memory. This creates a critical blind spot for most antivirus engines and endpoint detection and response (EDR) solutions, which are primarily designed to scan CPU processes and system RAM, not the dedicated memory and processing pipelines of a graphics card.
This attack highlights a broader trend in the cybersecurity landscape: the blending of legacy threats with cutting-edge techniques. USB worms like Retadup were a major problem over a decade ago but became less effective as operating systems improved their handling of removable media. However, attackers are gambling that complacency has set in and that many corporate environments have relaxed physical security policies, such as disabling USB ports. By pairing this simple, physical entry method with a complex, hardware-based evasion tactic, the attackers can bypass sophisticated network perimeter defenses. This hybrid strategy is particularly dangerous for environments that rely on air-gapped systems, such as industrial control or research facilities, where USB drives are a common method for transferring data. The use of non-traditional hardware components for malicious purposes is a clear signal that threat actors are actively exploring every possible avenue to avoid detection.
For security teams, CTOs, and developers, the emergence of CoffeeLoader serves as a critical reminder that a defense-in-depth strategy is non-negotiable. The threat model must account for both physical and digital vectors. This incident underscores the need to revisit and enforce strict policies on the use of removable media. Furthermore, it challenges security vendors and internal teams to expand their monitoring capabilities beyond the CPU and system memory. Endpoint security tools must evolve to gain visibility into GPU processes and other peripheral hardware. In the short term, security teams should focus on behavioral analysis and anomaly detection. Hunting for suspicious process chains, such as `rundll32.exe` executing an unfamiliar DLL from a temporary directory, can help identify the initial stages of this attack even if the final payload remains hidden on the GPU. This is a clear indicator that the next frontier of endpoint security will involve monitoring the entire hardware stack.
Why it matters
CoffeeLoader's use of GPU-based decryption is a significant evolution in malware evasion. This technique moves malicious operations off the CPU and out of system memory, bypassing traditional antivirus and memory scanning tools that are not equipped to inspect GPU computations, creating a major blind spot for defenders.
Business impact
The return of a USB-based worm combined with advanced evasion techniques increases the risk of breaches through physical vectors. This threatens not just standard corporate networks but also potentially air-gapped or industrial control systems, posing a serious operational and data security threat that bypasses perimeter defenses.
Tags
Related on Notifire
Related stories
Primary source: SecureList