5 Privacy
This chapter covers
- When GenAI privacy fails in practice.
- Four pillars of data protection in GenAI
- Practical steps to reduce risks in each pillar
- How deployment posture changes privacy risks
- What evidence regulators look for
When Elena Rodriguez filed an insurance claim with ElectraMotors, she expected ClaimAssist to use the photos, chat, and policy details needed to settle that claim. She did not expect the conversation to become training material or a permanent memory. Six months later, during a billing dispute, the chatbot greeted her by name and referred to a claim from ten years earlier. Then it added a rear collision that never happened.
For a privacy team, both failures matter. The old claim detail shows personal data traveling farther than Elena expected. The invented collision shows that a model can create personal data too: a false statement about a real person can still affect how she is treated.
Conventional privacy work starts by asking where the data sits, who can access it, why it was collected, and how it can be deleted. GenAI changes that map. Personal data can enter prompts and logs, retrieval indexes, fine tuning data, model or adapter parameters, stored outputs, vendor systems, and backups. The privacy breach may not be an unlocked database. It may be an answer that repeats a private detail, infers something sensitive, or fabricates a claim about a person.