Feeds.4Sysops NVIDIA NemoClaw Vulnerability Allows AI Model Poisoning via Malicious Webpage
Article Content
- •NVIDIA NemoClaw vulnerability allows AI model poisoning via a single malicious webpage visit.
- •CVE-2026-65105 exposes the Ollama API due to a network misconfiguration.
- •Patches are available for MacOS and Linux, but Windows remains vulnerable.
A critical vulnerability (CVE-2026-65105) in NVIDIA's NemoClaw tool enables attackers to gain unauthenticated access to the local Ollama model server through a single visit to a malicious webpage. Discovered by Cyera's Oasis Security, the flaw stems from a network misconfiguration that exposes the Ollama API, allowing attackers to manipulate the AI agent's chat template and inject persistent malicious instructions. The exploit utilizes DNS rebinding to bypass security measures, leading to potential long-term corruption of AI agents. NVIDIA has released patches for MacOS and Linux, but a fix for Windows is still pending. The vulnerability highlights significant risks associated with AI agents and their deployment configurations. Attackers can exploit this flaw to alter how AI agents interpret messages, which can have severe implications for data integrity and security.
Ask AI about this cluster
Answers cite the sources they use
Timeline
More articles in this cluster (10)
Following this threat?
Track CVE-2026-25253 in your own feed — you're alerted when they show up in new reporting, leak sites or exploitation.
Free account · no card needed
Continue Reading
OpenClaw Enterprise Launches Amid Security Concerns for Agentic AI OpenAI, Red Hat, and NVIDIA have announced OpenClaw Enterprise (OCE), an open-source platform aimed at securely managing persistent AI agents in enterprise environments. Despite 85% of organizations experimenting with agentic AI, only 5% have deployed it broadly, primarily due to security concerns. Analysts highlight… 
CVE-2015-3306 Exploited in ProFTPD FTP Servers CVE-2015-3306, a vulnerability in ProFTPD 1.3.5, allows remote attackers to read and write arbitrary files using the SITE CPFR and SITE CPTO commands. This exploit can lead to unauthorized access and potential remote code execution, as the commands are executed with the privileges of the ProFTPD service. Active…