Critical Ollama Memory Leak Vulnerability Exposes 300,000 Servers Globally

A critical vulnerability in Ollama (CVE-2026-7482, "Bleeding Llama") lets an attacker submit a malformed GGUF model to force out-of-bounds heap reads during tensor conversion, leaking memory contents (prompts, system instructions, environment variables, API keys) that can be preserved via float16→float32 conversion and exfiltrated by pushing the crafted model to a remote server; the issue affects versions prior to 0.17.1, has a CVSS score of 9.1, and organizations are advised to upgrade, remove public exposure, restrict access, rotate secrets, and assume possible compromise.