Steno leverages the power of Large Language Models (LLMs) to provide products and services that are custom-tailored to meet the highest standards of excellence. To achieve this, we must train one or more models using our customer’s data. We deeply understand the importance of this data to our customers. Content and other intellectual property are the lifeblood of the modern economy and must be safeguarded. We are committed to maintaining the highest standards of data protection and transparency. This document outlines the policies and tools we use to ensure your expectations around data privacy are met, as we provide you LLM-based products and services.
<aside> 🔐 Each customer's data and trained models are kept strictly separate. We never use any customer's data in any way, outside of the products & services we provide to you. We never share your data with anyone, apart from 3rd parties you have approved in writing.
</aside>
We clearly communicate how and where LLMs are used in our product, including our process for training models on customer data. We give our customers total control over which LLMs to use, while providing guidance on the pros and cons of each. We provide detailed information about our LLM providers and their privacy policies.
We exclusively partner with LLM providers who contractually commit to not copying or utilizing our customers' data for their own training purposes. This ensures that your proprietary information remains exclusively for your benefit. Customers can choose which LLM provider to use, if they have a preference, and we are happy to offer advice and guidance in this regard.
All model training on customer data occurs in isolated, secure environments. These environments are separated from our main systems and are subject to strict access controls.
We offer free guidance to all our customers, to help sanitize your data. Using various minimization techniques, you can help to ensure only the necessary information is used for effective model customization. We also offer premium data analysis and curation services, for those who are looking for additional protection and assurance.
Customers have total control over which data is used for model training. We provide intuitive interfaces for customers to select, review, and approve data sets for training.
We offer tools that allow customers to audit what data has been used to train their custom models, ensuring full transparency in the process.
Customers can set and enforce data retention policies for both their original data and the trained models. We offer options for prompt deletion of training data after model completion if desired.
All data, both at rest and in transit, is encrypted using industry-standard protocols. Our training environments have additional security measures to protect customer data during the model training process. Refer to our Security & Trust guide for more information.