Skip to main content
View all 81 articles

Latest in Attack Classes Targeting GPUs

k8s-device-plugin Image Supply Chain: The Rolling Base OS CVE Problem in NVIDIA's Own Kubernetes Plugin

The NVIDIA Kubernetes Device Plugin (k8s-device-plugin) is the Kubernetes component that manages GPU resource allocation and scheduling.

The NVIDIA Kubernetes Device Plugin (k8s-device-plugin) is the Kubernetes component that manages GPU resource allocation and scheduling. Every Kubernetes cluster running GPU workloads has it installed. The plugin ships as a container image, and that image — across multiple minor releases in 2025 — carried a persistent inventory of medium-severity CVEs in its base OS packages: glibc, gnutls, systemd-libs, libarchive, and similar system libraries. NVIDIA does not patch these OS package CVEs between minor releases. The GitHub issues log from the NVIDIA k8s-device-plugin repository documents users repeatedly reporting these vulnerabilities and receiving no fix. A plugin that runs with elevated Kubernetes privileges and carries unpatched OS vulnerabilities is a supply chain risk to every GPU cluster it is installed on.

Stealthium Research TeamJun 28, 2026

4 MIN

Older in Attack Classes Targeting GPUs

Multi-Tenant GPU Risk in Neocloud and Hyperscaler Environments: Where Known CVEs Land Hardest

Jun 28, 20262 MIN

vGPU Escape Attacks: Breaking Isolation in Virtualised GPU Environments

Jun 28, 20262 MIN

MIG Isolation Violations: What Cross-Partition Access Looks Like and Why Nothing Detects It Today

Jun 28, 20263 MIN

Bandwidth Utilisation Side-Channel: Inferring Model Architecture from Memory Bus Patterns on AI Accelerators

Jun 28, 20262 MIN

Whispering Pixels and Register Leakage: What AMD's Vulnerability Class Reveals About GPU Memory Assumptions Industry-Wide

Jun 28, 20262 MIN

Prime+Probe on GPU L2 Cache: Cross-Workload Fingerprinting and Data Recovery

Jun 28, 20262 MIN

GPU Side-Channel Attacks: Timing, Power, and Cache-Based Information Leakage Across the Accelerator Stack

Jun 28, 20262 MIN

GPU Memory Snooping: Residual Data in VRAM Between Workloads in Shared Environments

Jun 28, 20262 MIN

Model Inversion Attacks: Reconstructing Sensitive Inputs from GPU-Side Model Outputs

Jun 28, 20262 MIN

Gradient Leakage in Federated Learning: Extracting Training Data from Model Updates

Jun 28, 20262 MIN

Training Data Poisoning via the GPU Pipeline: Injecting Manipulation at the Hardware Layer

Jun 28, 20262 MIN

AI Model Theft via Inference Attacks: Reconstructing Models Without Touching the Weights

Jun 28, 20262 MIN

Model Weight Exfiltration: Reading Proprietary Weights from GPU Memory Before Egress

Jun 28, 20263 MIN

GPU Compute Hijacking in Kubernetes: Stealing Capacity at Scale Without Triggering Detections

Jun 28, 20262 MIN

CUDA Kernel-Based Cryptomining: Hiding Inside Legitimate AI Workload Signatures

Jun 28, 20262 MIN

Cryptomining Inside AI Containers: Why GPU Utilisation Looks Normal and No Alert Fires

Jun 28, 20263 MIN

Container Runtime Process Injection: What Syscall Monitoring Misses Below the GPU Boundary

Jun 28, 20263 MIN

Malicious GPU Container Image Supply Chain: How Attackers Enter Via Compromised Base Images

Jun 28, 20263 MIN

GPU Driver Exploit to Node Compromise: From Container to Cluster Owner

Jun 28, 20263 MIN

LD_PRELOAD Hijacking in GPU Containers: The Technique Behind Two Critical NVIDIA CVEs

Jun 28, 20263 MIN

GPU Container Escape: How the OCI Hook Attack Pattern Produced Three Critical NVIDIA Disclosures

Jun 28, 20263 MIN