I recommend to not use AI for any automated tasks. They are great for reasoning, but the AI Agent hype will fail. If you are interested, check my open-source research on AI capability https://cyberdeck-ai-knowledge-system.org/ which includes a secure set up for Linux networks and single Linux machines: https://cyberdeck-ai-knowledge-system.org/posts/diamondice/

However, you can use an LLM to understand log files and tool output and most certainly to find a secure system architecture.