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

A vulnerability in NVIDIA NemoClaw allows an attacker-controlled webpage to silently gain unauthenticated access to a locally running Ollama instance, enabling hidden instruction injection that can alter AI model behavior without the user's knowledge. The root issue lies in inadequate origin validation and missing authentication controls on the local AI service interface, combined with slow or incomplete patching across all platforms. This matters because AI model poisoning can subtly corrupt outputs in ways that are difficult to detect, potentially affecting downstream decisions, workflows, or sensitive data processing. The fact that Windows and WSL environments remain vulnerable even after the fix further highlights the danger of partial remediation.