AI Summary

University of Toronto hardware security researchers have created a new memory‑bit‑flipping method that outperforms prior GPU attacks. The technique, called GPUThor, bypasses the error‑correcting code (ECC) protections on enterprise Nvidia GPUs and can grant root access to the host system. GPUThor belongs to the Rowhammer class of attacks, which exploit charge leakage between densely packed RAM cells by repeatedly reading a memory row. By inducing controlled bit flips, the attack can be used for privilege escalation or manipulation of AI models.

AI summaries can be wrong sometimes—always verify important details using the source article.