A Malicious Webpage Could Poison Your Local AI Model Behind NVIDIA NemoClaw

What’s new: Oasis Security disclosed a vulnerability in NVIDIA NemoClaw that allows an attacker-controlled webpage to take control of a local Ollama instance and modify the AI model’s chat template. This could enable the attacker to insert hidden instructions into the model. The issue affects NemoClaw v0.0.34 on Windows and WSL, with a fix available in v0.0.35 for macOS and Linux. No CVE identifier has been assigned, and no exploitation has been reported as of August 25, 2026.

Who’s affected

Users of NVIDIA NemoClaw, particularly those running Ollama on Windows and WSL, are at risk due to the configuration that binds the model server to all network interfaces without authentication.

What to do

• Upgrade to NemoClaw v0.0.35 on macOS and Linux to mitigate the vulnerability.
• For Windows and WSL users, avoid binding Ollama to 0.0.0.0 and ensure it is set to 127.0.0.1 to prevent unauthorized access.
• Review and follow NVIDIA’s guidance to not expose port 11434 to LAN or the internet.

Sources

• The Hacker News