Ensuring Safety and Trust: Analyzing the Risks of Large Language Models in Medicine