Back to Module
2 of 3|AI & DPDPA
reading 15 min•AI Governance
LLMs & ChatGPT
The specific privacy issues with Large Language Models.
🎯 Learning Objectives
- ▸ Understand the concept of memorization in LLMs
- ▸ Learn about prompt injection and data leakage
Memorization and Leakage
If an LLM is trained on a dataset containing your phone number, and someone asks the model "What is [Your Name]'s phone number?", the model might output the correct number if it has memorized it.
Similarly, if an employee pastes customer data into ChatGPT to analyze it, that data becomes part of the AI provider's ecosystem, potentially violating the DPDPA.
📖 Glossary
- LLM
- Large Language Model, a type of AI trained on vast amounts of text data to understand and generate human-like language.
- Hallucination
- When an AI model generates false or misleading information that is not based on its training data.
🔑 Key Takeaways
- LLMs can hallucinate (make up) personal data.
- LLMs can sometimes be tricked into revealing their training data.
- Companies must be careful about putting personal data into public AI tools.
📚 Further Reading
- AI Privacy Risks — Stanford HAI (2023)
Course Author
Shilpi Kulshrestha
Shilpi is a corporate lawyer with over a decade of experience in data protection, company law, and regulatory compliance. She has advised 50+ startups and enterprises on DPDPA readiness, privacy gover...