NVIDIA has issued an urgent security bulletin urging customers using its Hopper HGX 8-GPU High-Performance Computing (HMC) systems to immediately install firmware updates addressing two critical vulnerabilities.
Released on February 28, 2025, the patches target flaws that could allow attackers to execute malicious code, escalate privileges, or cripple enterprise GPU infrastructure through denial-of-service (DoS) attacks.
The advisories highlight risks to AI data centers, cloud platforms, and scientific computing environments relying on these high-end accelerators.
The highest-risk vulnerability, tracked as CVE-2024-0114 resides in the Hopper HGX Management Controller (HMC)—a critical component overseeing GPU fleet operations.
Attackers with administrative access to a system’s Baseboard Management Controller (BMC) could exploit this flaw to gain unfettered HMC administrator privileges. NVIDIA’s advisory warns that successful exploitation enables:
The attack vector requires local network access but poses particular danger in shared infrastructure environments where BMC credentials might be exposed.
NVIDIA emphasizes that organizations using default BMC configurations face heightened risk.
A secondary vulnerability, CVE-2024-0141, affects the GPU vBIOS firmware governing low-level hardware operations.
Malicious actors with tenant-level GPU access could write malicious values to unsupported registers, forcing GPUs into unrecoverable failure states.
While rated medium severity, this flaw threatens cloud service providers by allowing rogue users to destabilize shared GPU resources.
The vBIOS exploit requires no physical access, making it viable in virtualized environments. NVIDIA notes that attacks would bypass traditional virtualization security layers, potentially crashing entire GPU partitions.
NVIDIA has released firmware version 1.6.0 to address both vulnerabilities, with backward compatibility for systems running earlier HMC builds. The company advises:
“These patches are non-optional for any organization using Hopper systems in production,” said Priya Balasubramanian, lead security engineer at NVIDIA’s Product Security Incident Response Team (PSIRT).
“The combination of code execution and hardware bricking capabilities creates unacceptable operational risks.”
The disclosure follows increased scrutiny of GPU security as AI workloads expand. Researchers warn that compromised accelerators could enable model theft, training data poisoning, or supply-chain attacks across ML platforms.
With over 80% of AI data centers utilizing NVIDIA GPUs, timely patching remains critical to global infrastructure integrity.
Collect Threat Intelligence on the Latest Malware and Phishing Attacks with ANY.RUN TI Lookup -> Try for free
Federal authorities have unveiled details of a sophisticated cybercrime operation targeting financial institutions across four…
A significant leak of internal chat logs from the Black Basta ransomware group has provided…
Security researchers at Socket have uncovered a sophisticated malware campaign targeting the Go ecosystem. The…
A sophisticated malware campaign has been uncovered, exploiting the growing popularity of Windows Packet Divert…
A recent physical penetration test conducted by cybersecurity firm Hackmosphere, revealed critical security flaws in…
A newly discovered malicious campaign dubbed "Desert Dexter" has infected approximately 900 victims across multiple…