vulnerability
A Malicious Webpage Could Poison Your Local AI Model Behind NVIDIA NemoClaw
High
Summary
A vulnerability in NVIDIA NemoClaw allows an attacker to poison an AI model by serving a malicious webpage that can inject hidden instructions into every conversation. The vulnerability stems from a misconfigured Ollama backend that exposes the API to all network interfaces, bypassing authentication and allowing an attacker to modify the model's chat template. While NVIDIA has released a fix, the vulnerability remains unpatched for many installations. The issue was previously identified and exploited against Paperclip, demonstrating a similar technique.
Summary written automatically in our own words from the original article, which belongs to its publisher and remains the reference. It may contain errors. Sources & data