Trustworthy AI Inference Systems: An Industry Research Perspective

Trustworthy AI inference systems need to provide customized inference to customers while protecting AI models with security mechanisms and protecting customer data at all times with privacy-enhancing technologies (PETs). This industry research perspective reviews the relationship between AI inference system trends and intellectual property and private data protection, surveys security and privacy building blocks, and focuses on the opportunities and challenges of combining trusted execution environments with cryptographic techniques to protect data in use. Future development requires joint attention from industry, academia, and government researchers.