Skip to main content
Emerging ThreatsMalware & Ransomware

Linux Cryptomining Malware Exploits Modified NVIDIA Library to Evade Detection

Rows of computer servers and networking equipment in a calm, neutral Linux data center.

"The threat actor also deployed a kernel rootkit to hide selected processes and files, an NVIDIA Management Library (NVML) wrapper to falsify GPU readings, and a separate backdoor named netns," Elastic Security Labs said in a report published last week.

SANSAR: a Rust orchestrator using TOR for command and control

Elastic Security Labs attributes a late‑October 2026 cryptomining campaign to a Linux‑focused operation tracked as REF3364 that centrally uses a Rust‑based toolkit called SANSAR. The orchestrator reaches an onion service (efxisnqldelyr5ra25oqfrsut54b34g26e7zrsiuupsoekjof2gho2id[.]onion) over TOR via Arti to receive commands, collect host and GPU information, and report miner status.

SANSAR supervises the mining processes and decides whether to launch a CPU or GPU miner based on instructions from the C2. It also performs ancillary tasks: launching or stopping miners, updating miner configuration, spawning interactive shells, managing connections, running an IPv4 TCP SYN traffic task against specified addresses and ports, and maintaining the NVML hook that falsifies GPU telemetry. As an anti‑debugging check, SANSAR reads TracerPid and exits if it is non‑zero.

NVML wrapper falsifies GPU telemetry and hides workload

The intrusion delivers a modified NVIDIA Management Library (NVML) wrapper that intercepts calls used by monitoring tools such as nvidia‑smi. The altered wrapper filters process information and reports GPU utilization as 0% while substituting temperature and power telemetry with randomized values between 36–46 °C and 100–200 watts, respectively.

Elastic warned that these changes effectively hide high GPU workloads from standard monitoring, allowing cryptomining to divert compute capacity from legitimate GPU tasks without raising straightforward alerts. The attacker also ran jobs invoking nvidia‑smi early in the chain to inventory GPU resources and exfiltrate that inventory to an external server.

acpi_thermal_mgr.ko kernel rootkit, initramfs persistence, and DKMS build

After obtaining root, the actor staged the toolkit under /opt/vadi_ and built a kernel module named acpi_thermal_mgr.ko through Dynamic Kernel Module Support (DKMS), then added an initramfs loader so the module and loader are stored in the boot image and can load again on subsequent boots. Security researcher Salim Bitam described this process in Elastic's report.

The rootkit uses Function Tracer (ftrace) to intercept system calls and conceal cryptomining activity by hiding matching processes, files, and socket entries. It does so by removing entries from directory listings or returning "No such file or directory" and "No such process" for requests that target hidden objects.

netns: an Ethereum‑assisted backdoor that resolves C2 via smart contract

A second notable payload is a backdoor called netns, registered as a systemd service. Netns retrieves a controller IPv4 address by sending requests to an embedded Ethereum smart contract (0x43eb291f56688c63c93e8a93c4591fc997cf6349). Once it resolves the address, netns establishes a plain TCP connection to that IPv4 on port 8080 and opens a reverse‑shell connection to the specified host and port in the C2 response.

Elastic also documented how the campaign used common techniques to gain and retain access: initial access via an exposed Ray job agent, submission of a Python reverse‑shell job that connected to a remote listener, and the addition of an SSH key to an account's authorized_keys file to establish persistence.

What this means for technologists, enterprise GPU operators, and procurement leaders

  • Technologists and security teams: verify whether GPU monitoring libraries (NVML or wrappers) have been modified and check for unexpected alterations to initramfs or DKMS‑built modules; be aware that tools like nvidia‑smi can be lied to by userland wrappers.
  • Enterprise GPU operators: review who can submit Ray jobs or otherwise schedule work on GPU clusters, and monitor for unusual exfiltration activity after job submissions and inventory calls.
  • Procurement and platform owners: require visibility into boot images and DKMS‑built modules and include integrity checks for boot‑time artifacts to detect modules staged into initramfs and stored on disk.

The operation packages familiar components — XMRig (kurt, version 6.25.0) for CPU mining, SRBMiner‑MULTI (tilki, version 3.6.9) for GPU mining, the lyrebird pluggable transport proxy for TOR, the acpi_thermal_mgr.ko rootkit, netns backdoor, and the sansar controller — into a coordinated toolset. Elastic noted Turkish language comments in the rootkit's source code but did not attribute the campaign to a named actor.

Elastic's direct advice echoes through the technical details: "Organizations expanding GPU infrastructure for AI and LLM workloads should check who can submit jobs and whether GPU monitoring libraries have been modified, as cryptojacking can divert compute capacity from legitimate workloads and increase operating costs," the report said. For operators who face rising demand for GPU cycles, that admonition is immediate and specific: the danger is not only lost cycles, but stealthy, boot‑persistent tooling that can survive reboots and fool common observability tools.

Original story