SANSAR Linux Cryptomining Malware Uses Modified NVIDIA Library to Hide GPU Activity
A new Linux-based cryptomining malware, SANSAR, uses modified NVIDIA libraries to conceal GPU activity, raising concerns for organizations utilizing GPU resources.

A threat activity cluster, identified as REF3364, has been linked to a Linux-focused cryptocurrency mining operation that manipulates GPU workload readings to evade detection. This operation employs a toolkit known as SANSAR, which orchestrates CPU and GPU miners while communicating through the TOR network.
The SANSAR toolkit, built with Rust, orchestrates the management of CPU and GPU miners, receiving commands via the TOR anonymity network. Observations of this activity began in late October 2026. According to Elastic Security Labs, the operation utilizes a kernel rootkit to conceal specific processes and files, alongside a modified NVIDIA Management Library (NVML) wrapper that falsifies GPU readings. A separate backdoor named netns is also part of the toolkit, which retrieves its command-and-control address from an Ethereum smart contract before establishing a direct connection to the server.
Initial access to the targeted systems was reportedly gained through an exposed Ray job agent. This vulnerability allowed the attacker to execute commands, including those running the NVIDIA System Management Interface (nvidia-smi), which captures and exfiltrates GPU inventory data. Following this, a Python reverse shell job connected to a remote listener, and an SSH key was added to the authorized_keys file to maintain persistent access to the system.
Once the attacker obtained root access, they staged the SANSAR toolkit in the /opt/vadi_ directory. They built the acpi_thermal_mgr kernel module using Dynamic Kernel Module Support (DKMS) and added an initramfs loader. An initramfs serves as a temporary root filesystem loaded into RAM during the early stages of booting, prior to switching to the main root filesystem. This setup ensures that the rootkit can reload upon system reboot.
A key feature of the SANSAR toolkit is its modified NVML wrapper, which filters process information and alters GPU telemetry to avoid detection. It hooks into NVML calls used by monitoring tools like nvidia-smi, reporting GPU utilization as 0%. Additionally, it replaces actual temperature and power readings with random values, making it difficult for monitoring systems to detect abnormal activity.
The acpi_thermal_mgr.ko kernel rootkit employs Function Tracer (ftrace) to obscure cryptomining activities on Linux systems. It hides processes, files, and socket entries by intercepting system calls related to directory listings, file access, and process termination signals. This allows it to remove records of hidden processes and files or block requests, returning messages indicating that no such file or process exists. The netns backdoor, registered as a systemd service, retrieves a controller IPv4 address through an embedded contract and establishes a reverse-shell connection to the designated host and port specified in the command-and-control response.
While the identity of the threat actor remains uncertain, Elastic Security Labs noted the presence of Turkish language comments within the rootkit's source code. Organizations expanding their GPU infrastructure, particularly for AI and large language model workloads, are urged to review access controls and ensure that GPU monitoring libraries remain unaltered to prevent cryptojacking, which can disrupt legitimate workloads and inflate operational costs.
- The specific identity of the threat actor behind the SANSAR operation is not confirmed.
- Details regarding the full extent of the affected systems or organizations remain unclear.
What it meansThis incident highlights the risks associated with unauthorized modifications to GPU management tools, which can lead to resource diversion and increased costs.
Ask IT this weekCan we confirm that our GPU monitoring libraries are intact and unmodified?Who has the authority to submit jobs to our GPU infrastructure?
DecisionEnsure regular audits of GPU access controls and monitoring systems.