security-course

Security, Privacy, and Consumer Protection

View the Project on GitHub noise-lab/security-course

What Do LLMs Remember About You?

1. Overview

Large Language Models (LLMs) like ChatGPT, Claude, and open-source systems such as Phoenix AI (UChicago’s internal LLM) have raised important questions about privacy. Models trained on massive datasets can memorize rare sequences — sometimes regurgitating sensitive information. In this activity, you’ll explore how inference and memorization risks play out in real-world systems, and investigate what settings are (or are not) available to users to protect their data.

We’ll look at real case studies and then test a few LLM interfaces to see what privacy tools are built in — and how transparent they really are.


2. Learning Objectives

By the end of this session, you should be able to:


3. Activity

Step 1: Case Study Discussion

Read the short summary of a real case study (provided by the instructor) where an LLM appeared to memorize or infer sensitive data. Examples may include:

In your group, discuss:

Step 2: Hands-On: Privacy Controls in LLMs (20–25 minutes)

Pick at least two LLM interfaces from the list below and try to answer the following questions by exploring the UI, settings, or documentation:

LLM Interfaces to Explore:

Each group should take notes on their findings and compare how different systems approach privacy and user control.


4. Discussion

As a class, we’ll talk about:

We’ll wrap up by considering: How can developers, institutions, or regulators create stronger norms and expectations around LLM privacy?