Privacy breach! Can confidential computing help generative AI take off more safely?
Large language models are trained by a massive amount of data laying around the world wide web. The training of these models is happening in an era where the topic of data privacy and security are talked about more than ever. Major data security and privacy concerns are on the table as the human-comparable skills of generative AI has taken off rapidly. How can enterprises keep pace with the trends of generative AI without compromising the security and privacy of their and users’ data?
Confidential computing is a technology that isolates sensitive data within the cloud into a protected part of the CPU. CPU is the part of a computer that processes inputs, stores data and outputs results. ChatGPT 3.5 gives a restaurant-related metaphor to describe how CPU works: “Imagine the CPU of a computer as a bustling kitchen in a restaurant. The inputs are the orders placed by hungry customers, the data is the fresh ingredients and recipes, and the outputs are the delicious meals served to the satisfied diners.” Without CPU, a computer does not work and without a protected CPU, sensitive data would leak.
Generative AI has the capacity to consume an entire company’s data and output innovative and unique results based on the data fed to it. Without confidential computing it would be difficult for companies to ensure that their sensitive data is protected and regulatory requirements – that are constantly evolving – are met. Confidential computing is gaining traction as the solution to avoid data security and privacy breaches: It allows enterprises to leverage from generative AI and keep up with industry trends without needing to constantly worry about, for example, intellectual property leakage in case of a breach.
Ayal Yogev in an article How Confidential Computing Could Secure Generative AI Adoption (TechCrunch.com 30.6.2023) says: “With confidential computing, data and IP are completely isolated from infrastructure owners and made only accessible to trusted applications running on trusted CPUs.” In addition, Yogev says that data security and privacy will be an indispensable part of confidential computing and enterprise leaders need to show it more light. For companies to adopt AI into their operations more safely, confidential computing would be the solution.