I’ll provide some clarification based on my perspective.
The llm’s that are used in n8n workflows are not trained by n8n. They were trainied by companies like openai
The LLM’s that you have in your n8n workflows can use whatever data you give them as further “training”. If you give gmail access to your LLM, it can access your gmail data to answer questions for the user. It doesn’t just grab all of your data in gmail, it only looks for what the user has asked for via the LLM.