PyTorch Archives • Daily CyberSecurity

AI Infrastructure at Risk: NVIDIA Fixes Critical 9.0 RCE Flaw in Apex Library (CVE-2025-33244)

NVIDIA GPU Security CUDA-Q Vulnerability NVIDIA Apex Vulnerability AI Infrastructure Security NVIDIA Cumulus Linux CVE-2025-33179 NVIDIA Arm divestment NVIDIA DGX Spark, CVE-2025-33187 NVIDIA Isaac-GROOT, Code Injection Megatron-LM Vulnerability, AI Code Injection NVIDIA China NVIDIA GPUs, Hardware Kill Switches NVIDIA Megatron-LM, LLM Vulnerabilities NVIDIA Base Command Manager - CVE-2024-0138

AI Infrastructure at Risk: NVIDIA Fixes Critical 9.0 RCE Flaw in Apex Library (CVE-2025-33244)

Do Son March 25, 2026

NVIDIA patches a critical 9.0 CVSS vulnerability (CVE-2025-33244) in the Apex library. Protect your AI models from RCE—upgrade to PyTorch 2.6+ now.

Hardware Autonomy: Meta’s “High Velocity” Roadmap to Deploy Four AI Chips by 2027 Meta MTIA roadmap

Hardware Autonomy: Meta’s “High Velocity” Roadmap to Deploy Four AI Chips by 2027

Do Son March 13, 2026

Meta shatters the industry standard with a 6-month chip cycle. Discover the MTIA 300-500 roadmap designed to slash AI costs and power 3 billion users by 2027.

Safety Broken: PyTorch “Safe” Mode Bypassed by Critical RCE Flaw PyTorch vulnerability, CVE-2025-32434 CVE-2026-24747

Safety Broken: PyTorch “Safe” Mode Bypassed by Critical RCE Flaw

Do Son January 29, 2026

PyTorch patches CVE-2026-24747 (CVSS 8.8). The "weights_only" safety flag failed, allowing RCE via malicious models. Update to v2.10.0 now.

The Desk-Side Revolution: NVIDIA’s DGX Spark Update Delivers 2.5× AI Speed Boost NVIDIA DGX Spark CES 2026, RTX 5090 AI collaboration

The Desk-Side Revolution: NVIDIA’s DGX Spark Update Delivers 2.5× AI Speed Boost

Do Son January 6, 2026

NVIDIA at CES 2026: DGX Spark gets a 2.5× speed boost, Nsight Copilot for offline CUDA coding, and 8× faster video gen for MacBook Pro users.

vLLM Flaw (CVE-2025-62164) Risks Remote Code Execution via Malicious Prompt Embeddings vLLM Vulnerability CVE-2026-22778 vLLM Memory Corruption, AI Inference RCE CVE-2025-29783 CVE-2025-32444 vLLM vulnerability vLLM, Remote Code Execution

vLLM Vulnerability CVE-2026-22778 vLLM Memory Corruption, AI Inference RCE CVE-2025-29783 CVE-2025-32444 vLLM vulnerability vLLM, Remote Code Execution

vLLM Flaw (CVE-2025-62164) Risks Remote Code Execution via Malicious Prompt Embeddings

Do Son November 24, 2025

A RCE flaw (CVE-2025-62164) in the vLLM inference engine allows attackers to crash servers or execute code by exploiting an out-of-bounds memory write via malicious prompt embeddings.

Critical Keras 3 RCE Flaw (CVE-2025-49655, CVSS 9.8) Allows Code Execution on Model Load Keras RCE, Insecure Deserialization

Critical Keras 3 RCE Flaw (CVE-2025-49655, CVSS 9.8) Allows Code Execution on Model Load

Do Son October 20, 2025

A Critical (CVSS 9.8) RCE flaw in Keras 3’s Torch backend (CVE-2025-49655) allows attackers to execute arbitrary code by loading a malicious model due to insecure deserialization.

Critical PyTorch Vulnerability CVE-2025-32434 Allows Remote Code Execution PyTorch vulnerability, CVE-2025-32434 CVE-2026-24747

Critical PyTorch Vulnerability CVE-2025-32434 Allows Remote Code Execution

Do Son April 21, 2025

CVE-2025-32434: Critical RCE vulnerability in PyTorch’s torch.load(weights_only=True) affects ≤2.5.1. Update to v2.6.0 to patch this major flaw