Using Exposed Ollama APIs to Find DeepSeek Models | UpGuard

**Executive Summary:** The UpGuard analysis found ~7,000 exposed Ollama API endpoints worldwide (a ~70% increase from prior scans), with ~700 instances running DeepSeek models; these unauthenticated APIs expose model metadata and enable potential data leakage, unauthorized model access/modification, and resource abuse, while observed tampering (models renamed to indicate exposure) suggests some interaction with exposed instances though most deployments appear to be hobbyist rather than enterprise.