GPUHammer demonstrates that Rowhammer bit flips are practical on GPU memories, specifically on GDDR6 in NVIDIA A6000 GPUs. By exploiting these vulnerabilities, attackers can significantly degrade the accuracy of machine learning models, highlighting a critical security concern for shared GPU environments.
Rowhammer attacks pose a significant threat by allowing malicious actors to manipulate AI models through a single bit flip, potentially compromising their integrity and security. This vulnerability highlights the need for enhanced protections in the development and deployment of AI systems.