Analysis Summary

NVIDIA has issued urgent security advisories for its Hopper HGX 8-GPU platforms, highlighting a critical privilege escalation vulnerability (CVE-2024-0114, CVSS 8.1) and a medium-severity vBIOS flaw (CVE-2024-0141, CVSS 6.8).

CVE-2024-0114, an authentication bypass in the HGX Management Controller (HMC), allows attackers with Baseboard Management Controller (BMC) access to escalate privileges, execute unauthorized code, manipulate AI workloads, and move laterally across GPU clusters. The flaw persists through reboots due to non-volatile storage. Meanwhile, CVE-2024-0141 affects GPU vBIOS, enabling unauthorized register writes that disrupt power management and memory controllers, leading to system crashes requiring manual intervention.

These vulnerabilities impact critical AI, ML, and supercomputing infrastructures, emphasizing the systemic risks in GPU-accelerated computing environments. Affected firmware versions include HGX HMC firmware up to 1.5.0 and all vBIOS versions prior to 1.6.0. NVIDIA urges administrators to mitigate risks by isolating BMC interfaces, enforcing strict authentication, applying firmware updates via nvfwupd, and restricting GPU passthrough permissions in virtualized environments. The ability to exploit HMC’s NVSwitch Fabric, which interconnects up to 256 GPUs, makes these flaws an attractive target for APTs aiming to compromise distributed AI training jobs or exfiltrate proprietary models.

While NVIDIA's Secure Firmware Update framework cryptographically signs firmware updates to prevent tampering, the delayed disclosure of CVE-2024-0114—discovered in December 2023 but patched only in 2025—highlights the challenge of securing complex hardware/software integrations. As reliance on high-performance GPU clusters grows, enterprises must adopt continuous firmware monitoring and hardware-rooted Zero Trust architectures to mitigate supply chain and firmware-level threats effectively.

Impact

• Code Execution
• Denial of Service
• Privilege Escalation
• Information Disclosure

Indicators of Compromise

CVE

• CVE-2024-0114

• CVE-2024-0141

Affected Vendors

Remediation

• Upgrade HMC firmware to version 1.6.0+ using NVIDIA’s nvfwupd utility.
• Update to vBIOS 1.6.0+ to mitigate unauthorized register writes.
• Enforce strict network segmentation for IPMI/Redfish endpoints.
• Implement certificate-based authentication to prevent unauthorized access.
• Limit BMC access to trusted administrators and secure management networks.
• Regularly audit HMC non-volatile storage for unauthorized modifications.
• Restrict GPU passthrough privileges in virtualized environments to prevent tenant exploitation.
• Disable direct hardware register access for untrusted workloads.
• Track system logs for unexpected GPU crashes or power instability requiring resets.
• Continuously verify access requests, especially for HMC/BMC and GPU firmware updates.
• Keep all firmware up to date and conduct periodic vulnerability assessments.
• Restrict inter-node communication and enforce cryptographic integrity for firmware images.