LLMOps

Chapter 8. Governance: Monitoring, Privacy, and Security

We hear the words privacy and security all the time, especially when talking about technology, and many people assume they’re the same thing. In fact, they’re very different concepts. Privacy is about control over your personal information—who gets to know what about you. Security, on the other hand, is about protecting that information from being stolen, leaked, or accessed without permission. They overlap, for sure, but understanding the difference becomes really critical when we talk about LLMs, because these models expose both privacy and security risks in ways no one has ever dealt with before.

Today, privacy is more important than ever. With AI, and especially LLMs, being integrated so seamlessly into so many products and services, it’s hard to keep tabs on what is still private and what isn’t. One major concern is that chat interfaces like ChatGPT, Gemini, and Claude are being adopted as easy-to-use search services, and their interactions can seem humanlike, potentially leading users to reveal more than they should. Robust cybersecurity has become a must-have for all AI and ML companies.

In June 2023, a New York law firm, Levidow, Levidow, and Oberman, was fined by a jury for using fake legal cases manufactured by ChatGPT in its research for an aviation injury claim. The media spent days discussing the unreliability of LLMs and lack of trust in the information they provide. Another serious issue is the need to educate ...