Description: America-based microchip company Nividia was recently hit by a cyber attack. The company's internal systems including email and developer tools were completely compromised. As for now, no data leak has been reported and the company has shut down its systems to contain the attack.

Description: A critical flaw, CVE-2024-0132, in NVIDIA’s Container Toolkit has remained exploitable due to an incomplete patch, leaving AI infrastructure and sensitive data at risk. This vulnerability, coupled with a newly found Docker DoS flaw on Linux, endangers systems by potentially letting attackers steal AI models, disrupt operations, or execute DoS attacks. The initial patch provided by NVIDIA was later found to be inadequate for versions 1.17.3 and below, and conditionally for 1.17.4, making them susceptible to breaches and command execution with root privileges. The Docker issue can result in CPU usage spikes and SSH access loss, disrupting services and rendering systems unresponsive. The exposure primarily affects organizations using these tools for AI or cloud workloads, including industries like healthcare, finance, and autonomous systems.

Description: Several Nvidia-owned web domains, including an events page, were hijacked to show explicit and AI-generated content in a spam campaign. The page was defaced and over 62,000 AI-generated articles were posted, containing incorrect or incomplete information about popular search topics. The attack also targeted NPR, Stanford, and US Government sites, including a domain belonging to the US Department of Health and Human Services (HHS). The links in the pages directed to a 'nonsense SEO spam page' stocks.wowlazy[.]com.

Description: NVIDIA released emergency security patches addressing two high-severity vulnerabilities in their Megatron LM large language model framework. These vulnerabilities, CVE-2025-23264 and CVE-2025-23265, affect all versions prior to 0.12.0. Exploitation of these flaws could allow attackers to execute code, escalate privileges, access data, and tamper with AI models through malicious file injection. The vulnerabilities were discovered by security researchers Yu Rong and Hao Fan and classified under CWE-94 with a CVSS v3.1 base score of 7.8. NVIDIA urges immediate updates to version 0.12.1 or later to mitigate these risks.

Description: Nvidia advised customers to ensure mitigations against Rowhammer attacks after researchers found one of its workstation-grade GPUs is susceptible. The advisory noted that researchers at the University of Toronto demonstrated a successful Rowhammer exploitation on a NVIDIA A6000 GPU with GDDR6 memory where System-Level ECC was not enabled. The company recommended customers ensure System-Level ECC is enabled on many models in its Blackwell, Ada, Hopper, Ampere, Jetson, Turing, and Volta products.