Cybersecurity News

Critical NVIDIA DGX Spark Flaws Allow Malicious Code Execution and Denial-of-Service Attacks

NVIDIA disclosed 14 vulnerabilities in its DGX Spark GB10 AI workstation on November 25, 2025, affecting all DGX OS versions before OTA0.

These flaws, mainly in the SROOT firmware and hardware resources, enable local attackers with privileged access to bypass protections, leading to remote code execution, data tampering, information disclosure, denial-of-service, and privilege escalation.

The DGX Spark, a compact Grace Blackwell-powered system for AI model training and inference, faces significant risks in enterprise environments that handle sensitive machine learning data.​

Critical issues like CVE-2025-33187 (CVSS 9.3, CWE-269) allow attackers to access SoC-protected areas via SROOT, potentially compromising the entire system through arbitrary code execution.

Similarly, CVE-2025-33188 (CVSS 8.0) permits tampering with hardware control, while out-of-bounds writes in CVE-2025-33189 (CVSS 7.8, CWE-787) and CVE-2025-33190 (CVSS 6.7) enable memory corruption for code execution or DoS.

Vulnerability Summary

Lower-severity flaws include invalid memory reads (CVE-2025-33191, CVSS 5.7, CWE-20), arbitrary memory reads (CVE-2025-33192, CWE-690), integrity check failures (CVE-2025-33193, CWE-354), and input mishandling (CVE-2025-33194, CWE-180), all of which risk DoS or leaks.

Additional risks from buffer issues (CVE-2025-33195, CWE-119), resource reuse (CVE-2025-33196/33198/33200, CWE-226), NULL pointer derefs (CVE-2025-33197, CWE-476), and control flow errors (CVE-2025-33199, CWE-670) further expose systems.

CVE IDVectorBase ScoreSeverityCWEKey Impacts
CVE-2025-33187AV:L/AC:L/PR:N/UI:N/S:C/C:H/I:H/A:H9.3Critical269Code exec, info disclosure, tampering
CVE-2025-33188AV:L/AC:L/PR:N/UI:N/S:U/C:L/I:H/A:H8.0High269Disclosure, tampering, DoS
CVE-2025-33189AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H7.8High787Code exec, tampering, DoS
CVE-2025-33190AV:L/AC:L/PR:H/UI:N/S:U/C:H/I:H/A:H6.7Medium787Code exec, tampering, DoS
CVE-2025-33191AV:L/AC:L/PR:N/UI:N/S:C/C:L/I:N/A:L5.7Medium20DoS
CVE-2025-33192AV:L/AC:L/PR:N/UI:N/S:C/C:L/I:N/A:L5.7Medium690Code exec, DoS, disclosure
CVE-2025-33193AV:L/AC:L/PR:N/UI:N/S:C/C:L/I:N/A:L5.7Medium354Code exec, DoS, disclosure
CVE-2025-33194AV:L/AC:L/PR:N/UI:N/S:C/C:L/I:N/A:L5.7Medium180Disclosure, DoS
CVE-2025-33195AV:L/AC:L/PR:L/UI:N/S:U/C:N/I:L/A:L4.4Medium119Tampering, DoS, escalation
CVE-2025-33196AV:L/AC:L/PR:H/UI:N/S:U/C:H/I:N/A:N4.4Medium226Disclosure
CVE-2025-33197AV:L/AC:L/PR:N/UI:N/S:C/C:N/I:N/A:L4.3Medium476Code exec, DoS
CVE-2025-33198AV:L/AC:L/PR:L/UI:N/S:U/C:L/I:N/A:N3.3Low226Disclosure
CVE-2025-33199AV:L/AC:L/PR:H/UI:N/S:C/C:N/I:L/A:N3.2Low670Tampering
CVE-2025-33200AV:L/AC:L/PR:H/UI:N/S:U/C:L/I:N/A:N2.3Low226Disclosure

Mitigation Steps

NVIDIA urges immediate upgrade to DGX OS OTA0, available from the DGX Spark product page and NVIDIA Product Security site.

The flaws were found by NVIDIA’s Offensive Security Research team, with no known exploits in the wild.

Organizations using DGX Spark for AI workloads should prioritize patching to safeguard high-value models and data from local threats.

Varshini

Varshini is a Cyber Security expert in Threat Analysis, Vulnerability Assessment, and Research. Passionate about staying ahead of emerging Threats and Technologies..

Recent Posts

Burp Suite Supercharges Its Scanning Capabilities With React2Shell Vulnerability Detection

PortSwigger has leveled up Burp Suite's scanning arsenal with the latest Active Scan++ extension, version…

4 months ago

Malicious MCP Servers Enable New Prompt Injection Attack To Drain Resources

Unit 42 researchers at Palo Alto Networks exposed serious flaws in the Model Context Protocol…

4 months ago

Law Enforcement Detains Hackers Equipped With Specialized Flipper Hacking Tools

Polish police have arrested three Ukrainian men traveling through Europe and seized a cache of…

4 months ago

Google Unveils 10 New Gemini-Powered AI Features For Chrome

Google has launched its most significant Chrome update ever, embedding Gemini AI across the browser…

4 months ago

CISA Alerts On Actively Exploited Buffer Overflow Flaw In D-Link Routers

Attackers exploit this vulnerability through the router's web interface components, specifically "cgibin" and "hnap_main," by…

4 months ago

Over 500 Apache Tika Toolkit Instances Exposed To Critical XXE Vulnerability

Security researchers have uncovered a severe flaw in Apache Tika, a popular open-source toolkit for…

4 months ago