Privacy first analytics with Crux

Atharva Padhye
October 2, 2023

Rising Privacy Concerns with LLM Applications

As an enterprise leader, data privacy is of utmost importance to your business. As you explore new use cases of Large Language Models (LLMs), you may have several concerns:

  • "Will GPT or other LLMs get access to my internal data when I explore use cases?"
  • "Will a 3rd party application use my data to train their models?"
  • "How do I derisk my organization from the privacy breaches of the LLM?"

At Crux, we thoroughly understand these concerns and have strived to address them from Day 0. Crux's privacy-first foundation and enterprise-grade security ensure that you can focus on exploring and executing the use cases that matter while the rest is taken care of.

This is one of the reasons why enterprises in sectors with the strictest privacy regulations (e.g. Pharmaceutical Manufacturing) have chosen Crux to redefine their customer-facing analytics.

How Crux maintains data privacy

Crux only uses the Metadata, and not the actual data

  • The Crux engine works in a way where only the metadata (table names, column names, value type, and column description) are required to successfully generate an SQL query
  • This SQL query is then run over your database or data warehouse to generate the final output
Meta Data

No data ever leaves your server

  • Crux is entirely deployed on-premise to ensure that all your data is always as it is, without ever leaving your server
  • We make sure that users have to use the designated SSH key (only shared with selective users in your organisation) to get access to any data

Your organization is safeguarded from LLM's privacy breaches

  • Most LLMs (OpenAI's GPT, Google's PaLM) have publicly disclosed that they don't store and use user prompts for any kind of training
  • However, in case any of these LLMs run into trouble, your organization must be protected from all compliance issues
  • Hence, we provide an option to fine-tune any open-source LLM of your choice (LLama, Claude, etc.) and host it entirely on your premise to derisk your organization
LLMs

You can choose who has access

  • Our configurable access control mechanism at a user level, function level, geography level, and organization level makes it possible to define the extent of data access that every user within your organization has
LLMs


Takeaways

The realm of Generative AI is new to the world and all of us are still figuring out how it fits into our lives. With these checks and balances in place, our partners can rest assured that their data is safe and they can shift their focus to innovation & core product development.

We would love to hear from you about how we can improve our privacy measures.

in this article