Ollama drama as 'easy-to-exploit' critical flaw found in open source AI server

A critical remote code execution vulnerability (CVE-2024-37032, "Probllama") in Ollama's /api/pull endpoint allows attackers to supply malicious model manifests with path traversal payloads, enabling file corruption, arbitrary file read, and RCE—particularly dangerous for Docker installs running the server as root and listening on 0.0.0.0. The flaw was patched in version 0.1.34, but researchers found more than 1,000 internet-exposed vulnerable instances; recommended mitigations are to upgrade, avoid exposing the server publicly, and enforce authentication/reverse-proxy protections.