Exposed Ollama Servers: Security Risks of Publicly Accessible LLM Infrastructure

This report warns that thousands of self-hosted Ollama LLM servers were found exposed to the internet, enabling attackers to enumerate models, submit prompts (including prompt-injection attempts), and abuse GPU/compute resources; it details the risks and provides practical mitigations—restrict binding interfaces, firewall/security group rules, private network deployment, authentication layers, monitoring, and asset inventorying—and notes Indusface WAS can help detect such exposures.