NVIDIA DGX Spark CVE-2026-24263

Exploitation requires local access to the DGX Spark system with high-privilege credentials (CVSS PR:H, AV:L). The attacker must have the ability to interact with or invoke the vulnerable firmware code path, which implies either physical presence or remote administrative access (e.g., BMC/IPMI, SSH as root, or equivalent firmware management interface). Low attack complexity (AC:L) indicates no race condition or specialized environmental setup is needed once privilege is obtained. Exploitation is limited by: (1) the requirement for high-privilege local access, ruling out remote unauthenticated exploitation; (2) the specialized hardware platform - only DGX Spark systems are in scope; and (3) the absence of user interaction means no phishing vector is available to an external attacker without prior access. The specific firmware feature or interface that triggers the NULL dereference is not identified in the available data.

The CVSS 3.1 base score of 8.2 (High) with vector AV:L/AC:L/PR:H/UI:N/S:C/C:H/I:H/A:H reflects a genuine but access-gated threat. The most significant mitigation is the prerequisite of local, high-privilege access (PR:H, AV:L), which substantially narrows the exploitable population to insiders, already-compromised privileged accounts, or attackers who have achieved prior privilege escalation. No EPSS data is included in the provided intelligence, and the vulnerability is not listed in CISA KEV, suggesting no confirmed widespread exploitation. The absence of a known POC further reduces near-term automated exploitation risk. However, the S:C scope change combined with all-High C/I/A impacts means that for any attacker who does satisfy the access prerequisite, the consequences are severe and extend beyond the firmware layer into the broader system. In AI/ML infrastructure environments where DGX Spark systems are shared among multiple users or remotely administered, the insider threat surface is non-trivial.

An attacker with privileged local access to a DGX Spark system - such as a malicious insider, a compromised administrative account, or an attacker who has already escalated privileges via a separate vulnerability - crafts or triggers a specific firmware code path that dereferences a NULL pointer. The resulting memory fault is leveraged to corrupt firmware execution flow, enabling arbitrary code execution within the firmware's trust boundary and potentially escaping into OS or hypervisor layers, achieving persistence or further privilege escalation. No public proof-of-concept has been identified at time of analysis.