Title: New GPUBreach Attack Exposes GPUs to System Takeover Risks
A groundbreaking cybersecurity threat, known as GPUBreach, has emerged, allowing attackers to compromise systems by exploiting vulnerabilities in GPU memory. Developed by researchers at the University of Toronto, this attack utilizes a method called Rowhammer to manipulate GDDR6 memories, enabling unauthorized access and potential system takeover.
The attack primarily affects users of NVIDIA GPUs, particularly the RTX A6000 model, which is widely used in AI development. By corrupting GPU page tables, attackers can escalate privileges from unprivileged CUDA kernels to gain full control of the system. Notably, this can occur even with existing memory protections in place.
To safeguard against GPUBreach, users should ensure their systems are updated, activate Error Correcting Code (ECC) memory where possible, and remain vigilant regarding security updates from GPU manufacturers.
Risk Level: High – The potential for complete system compromise poses a significant threat to users and organizations alike.
Source: View Original Report
