<ul data-eligibleForWebStory="true"><li>Ollama is an open-source toolkit for large language models, but an unauthenticated access vulnerability on the default port 11434 can expose sensitive API endpoints to the public internet.</li><li>The vulnerability allows attackers to access private models, cause resource abuse, or even execute remote code, posing serious risks like data theft and server compromise.</li><li>Changing Ollama's default settings to enable remote access without authentication can lead to significant security flaws, including the CVE-2024-37032 path traversal issue for remote code execution.</li><li>Mitigation steps include restricting public access, implementing access control via firewall rules or a reverse proxy like NGINX with basic auth, and updating to fixed versions like v0.1.34.</li><li>All Ollama versions are vulnerable if exposed to 0.0.0.0 without authentication, with related flaws addressed in v0.1.34. The issue has been observed being exploited in real-world scenarios.</li></ul>

Ollama Exposed: Unauthenticated Access Vulnerability Could Leak Your LLM Models

Discover more