A Guide for CISOs and AI Transformation Leaders
The advent of Large Language Models (LLMs), such as ChatGPT, Bard, and even open-source foundation models, has ushered in an era of unprecedented possibilities. These models can enable businesses for a broad range of tasks, from code generation to empowering your next top-performing customer service agent. But with this great power comes a great challenge when it comes to the enterprise: protection and ownership of proprietary enterprise data.
Enterprises often hesitate to utilize their most important data with hosted LLMs in the cloud due to the risk of exposing sensitive information. This concern is about data ownership rather than the security guarantees of LLM providers or the CSPs that host the models. As such, both enterprises and Generative AI providers will benefit from enabling the enterprise to retain ownership of plain-text data when using hosted foundation models to maximize the adoption of Generative AI and drive value.
Protopia AI has been working to solve this problem for the last three years – not just for Generative AI but for any data used in ML and AI. The Protopia AI team has built Stained Glass Transform™ (SGT), using patented technology designed to enable customers to safely and more confidently use any ML model while retaining ownership of their sensitive plain-text data.
How Stained Glass Transform Works
SGT converts raw/plain-text data into a Randomized Re-Representation that retains the information necessary for the model consuming it. This means that plain-text data does not appear in any queries, context, or fine-tuning data. Randomized Re-Representations of data are not, and can not be, reversed back to the original plain-text format when being used by the ML model on the provider side, which means the enterprise client retains ownership of plain-text information at all times. Additionally, Randomized Re-Representation can be used together with any encryption methodology; bring your own key or otherwise.
In a recent customer use case, Protopia AI’s SGT was successfully used to create Randomized Re-representations of log data, and the Stained Glass output was used to fine-tune a large language model for a classification task. The result? The transformed data – entirely indecipherable to any person or machine that had access to it – was used in the transformed format to fine-tune the LLM with virtually identical performance to what would have been achieved with plain-text data.
Existing Use-Case of Stained Glass Transform™ with LLMs
Randomized Re-Representations enable teams to use ML to analyze the underlying information without needing to access it in identifiable form.
Curious about the mathematical transformations that power Protopia AI’s technology? Explore our whitepaper on Foundational Data Protection for Enterprise LLMs for an in-depth understanding of the underlying magic.
Applications in Generative AI and LLMs
Stained Glass Transform is a game-changer when querying an already-trained LLM and when fine-tuning models for new tasks. In the case of querying, SGT retains ownership of plain-text queries and context by transforming the plain-text into randomized stochastic embeddings instead of deterministic embeddings. In the realm of fine-tuning, SGT protects the plain-text training data. Using SGT allows enterprises to query and/or fine-tune hosted models without exposing plain-text data.
Stained Glass isn’t Only For Text Data
Stained Glass Transform™ isn’t restricted to text data. Protopia AI has successfully demonstrated the applicability of Stained Glass Transform to visual data and structured or semi-structured data. From protecting trade secrets in the manufacturing process to enabling real-time face recognition for the U.S. Navy, SGT proves its versatility and efficiency in different use cases using AI.
US Navy Face-Recognition Use Case
A Giant Leap for Generative AI and LLMs
With the introduction of Protopia AI’s Stained Glass Transform™, the adoption of Generative AI and LLMs by enterprises is set to accelerate. By ensuring data ownership and privacy, SGT removes one of the biggest hurdles to the large-scale adoption of these models. It enables enterprises to harness the power of AI without the risk of data exposure, thus opening up a new world of possibilities in the field of AI.
AWS recently selected Protopia AI for its Generative AI Accelerator. The collaboration is geared to meet a crucial demand in the GenAI ecosystem, and Protopia AI is excited to play a pivotal role in facilitating the secure use of more enterprise data with generative AI at scale.
- Want to learn more about how Stained Glass Transform™ works for Generative AI? Request our Whitepaper on Foundational Data Protection for Enterprise
- If you want to get into more details. Set up a Free Consultation. Our team is excited to collaborate with you and solve your your unique requirements.