Analyze the security risks of local large model deployment tools like Ollama, focusing on API exposure, prompt extraction, and model information leakage. Provide a step-by-step guide for security testing, including API probing techniques and defense measures against prompt injection attacks. Also suggest best practices for secure configuration of local model deployment.