Building Trust Through Private and Verifiable AI

AI has transformed how people learn, work and live - automating complex tasks and extracting insight from massive datasets. But most powerful AI today (especially large language models) runs on server-class hardware, which typically means user prompts and context must be visible to the service provider to be processed. While acceptable for some cases, it is still challenging with highly sensitive data where users expect similar protections as end-to-end encryption. Private Verifiable Compute (Project PVC) is an open source project that can enable users to initiate a request to a private and verifiable environment for context-aware AI processing with sensitive data, where no one, including service providers, can access them. With PVC in the cloud environment, it unleashes full potentials of AI hardware in the data center for complex AI tasks, such as large language models (LLMs), generative AI and beyond, while guaranteeing user privacy and verifiable transparency.