Privacy & security

LLMs for Data Analysis in R

R/Medicine 2026

Privacy & Security

Before we wrap up, important to discuss privacy and security — especially relevant for those working with protected health data.

Model providers vs. LLMs

• Provider: The company that hosts the model (Anthropic, OpenAI, Google, etc.).

• LLM: The actual model that generates responses to your queries.

• LLMs are stateless. They have no memory of your prior requests.

• However, the model provider may log and store your requests.

The model doesn’t remember you, but the provider’s servers might.

lets walk through the flow of data when you have a convo, you really only see this it looks liek you send a message to anthropic or whoever, and you get back a response

the model provider is responsible for routing your message to the actual model

remember, LLMs are stateless. so the llm doesn’t store your request or any data. it just takes in your request and gives a response the model provider handles giving that response back to you

the nuance is that the model provider might do something with your data, and this can vary by rpovider and by agreement you have with the provider.

model training, third party, storage

when you use something like posit assistant, this diagram doesn’t change much again, teh llm itself isn’t going to retain any of your data. it doesn’t automatcaly “learn” or anything.

and posit doesn’t store your data unless two things are true

You* need to trust your model provider

* or more likely, your organization

• Know what your model provider does with your data

• ellmer, querychat, shinychat, etc. will never store your data anywhere. They pass the request directly to your specified model provider.

• Posit Assistant will only store your data if a) you use Posit as your model provider and b) you opt-in.

The good news

Your organization can work out a zero data retention (ZDR), HIPAA-compliant agreement, or other arrangements with model providers. These arrangements are common.

Learn more: https://posit.co/blog/trust-llm-tools/

Learn more

These are the Posit packages for working with AI. We’ve focused on ellmer and shinychat today, but there are others for different use cases.

Thank you!