GPU Vulnerabilities Pose Cybersecurity Risk to Intelligent Transportation Systems

The increasing reliance on video analytics and artificial intelligence is transforming modern transportation systems, but a critical security vulnerability remains largely unaddressed. Sefatun-Noor Puspa and Mashrur Chowdhury, from Clemson University, along with their colleagues, demonstrate that graphics processing units (GPUs), essential for these advanced applications, represent a significant blind spot in transportation cybersecurity. Their research reveals that GPUs are susceptible to stealthy attacks, such as unauthorized cryptocurrency mining, which can severely degrade performance and compromise the reliability of critical systems like autonomous vehicles and roadside infrastructure. By conducting a case study with a video processing pipeline, the team shows that malicious background processes can reduce frame rates by half and increase power consumption by up to 90 percent, highlighting the urgent need for improved GPU monitoring and the development of effective detection strategies. They further present a replicable framework, utilising on-device telemetry, to identify and mitigate GPU misuse, raising awareness of these observability gaps within intelligent transportation systems.

These GPUs provide high-throughput and energy-efficient computing for tasks such as sensor fusion and roadside video analytics. Despite their increasing importance, GPUs represent one of the most unmonitored components in terms of security, leaving them vulnerable to both cyber and hardware attacks, including unauthorized cryptocurrency mining. This research highlights GPU security as a critical blind spot in transportation cybersecurity, demonstrating a significant and often overlooked risk within the infrastructure.

GPU Monitoring Detects Perception System Attacks

This research investigates the potential security vulnerabilities of GPUs in autonomous vehicles, specifically focusing on how malicious activity, like cryptojacking, can impact perception systems and ultimately vehicle safety. The study reveals that autonomous vehicles heavily rely on GPUs for computationally intensive tasks such as processing data from cameras, LiDAR, and radar, and running perception algorithms like object detection and segmentation. These systems are increasingly vulnerable to attacks, including cryptojacking, which could potentially manipulate perception data. Even seemingly benign attacks can degrade GPU performance, leading to reduced frame rates, increased latency, and failures in the perception pipeline, potentially causing accidents.

The research demonstrates that GPUs present a unique attack surface due to their complex architecture, resource demands, and increasing connectivity. The team found that even moderate GPU load from cryptojacking can significantly impact the performance of perception algorithms, increasing latency and reducing accuracy. Perception systems are particularly vulnerable because they require real-time processing and are sensitive to even small performance fluctuations. The authors argue for the importance of monitoring GPU performance metrics, including utilization, temperature, and memory usage, to detect malicious activity and prevent performance degradation.

This work establishes that GPU security is critical for autonomous vehicle safety, requiring protection of both software and underlying hardware. Real-time monitoring of GPU performance is essential for detecting malicious activity early and preventing safety hazards. A multi-layered security approach, including hardware-level security features, software-based intrusion detection, and robust monitoring systems, is necessary. Further research is needed to develop more effective GPU security mechanisms and intrusion detection techniques for autonomous vehicles.

GPU Mining Stealthily Degrades Transport Safety Systems

Graphics processing units (GPUs) are increasingly vital for intelligent transportation systems, powering applications like sensor fusion and video analytics, yet remain surprisingly vulnerable to security threats. Recent research highlights a critical blind spot in transportation cybersecurity: the potential for unauthorized use of GPU resources by “crypto miners”, programs that solve complex mathematical problems to generate cryptocurrency. This study demonstrates how such stealthy miners can significantly degrade GPU performance, impacting the reliability of safety-critical applications. The research team observed a substantial reduction in processing speed, a 50 percent drop in frame rate, when a crypto miner ran alongside a standard video processing workload.

Simultaneously, power consumption increased by as much as 90 percent, demonstrating that the miner aggressively consumes available resources. This performance hit isn’t merely a matter of inconvenience; it directly affects the ability of transportation systems to process real-time camera input, potentially leading to missed detections or delayed responses to hazards. For example, a vehicle relying on video analytics for pedestrian detection might fail to react quickly enough if the frame rate is halved. Importantly, traditional system checks may not detect this problem, as the GPU remains active, creating a silent risk to both performance and functional safety.

The research team successfully developed and tested machine learning models capable of detecting the presence of a crypto miner by analyzing GPU telemetry data, metrics like power usage, memory consumption, and processing throughput. All tested models, including random forests, gradient boosting, linear regression, and neural networks, achieved 100 percent accuracy in distinguishing between normal workloads and mining activity. This high level of detection accuracy suggests that even limited access to on-device metrics can provide a robust defense against malicious GPU activity, without requiring modifications to system drivers or additional hardware. The findings underscore the urgent need for improved GPU observability within intelligent transportation systems, and demonstrate the feasibility of building a lightweight, effective security layer to protect critical infrastructure from covert resource hijacking. The ability to reliably detect and mitigate these threats is crucial for ensuring the safety and reliability of future transportation technologies.

GPU Misuse Impacts Autonomous System Safety

This research highlights a critical and often overlooked security vulnerability in intelligent transportation systems: the potential for misuse of graphics processing units (GPUs). The study demonstrates that unauthorized activity, such as cryptocurrency mining, running in the background can significantly degrade GPU performance, reducing frame rates by as much as 50 percent and increasing power consumption by up to 90 percent, without triggering conventional security alerts. This performance reduction poses a safety risk by subtly compromising the responsiveness and reliability of AI-powered systems used in autonomous vehicles and roadside infrastructure. To address this gap, the authors developed and tested a framework for detecting GPU misuse through on-device telemetry, achieving high accuracy with lightweight classifiers trained on GPU performance metrics. While acknowledging limitations in current detection methods, this work establishes a replicable approach for monitoring GPU runtime integrity. The authors emphasize the need for increased attention to GPU security, advocating for investment in GPU-aware monitoring tools, secure update channels, and the establishment of behavioral baselines for edge AI workloads to ensure both safety and resilience as transportation systems become increasingly reliant on GPU-accelerated computing.