Data Security, Professional Perspective - Regulation of AI Foundation Models

Artificial intelligence (AI) providers rely on foundation models that are pre-trained on massive amounts of data to operate their AI systems. Only a limited number of companies have the tools and resources to train foundation models, which serve as the bedrock for companies looking to provide AI solutions in the marketplace. Foundation models are often not trained to perform specific tasks. Rather, they are general-use models that companies can customize through fine-tuning to perform particular functions.

Today, AI service providers can integrate foundation models into new and existing products and services, such as search engines, productivity software, financial tools, social media, health care, robotics, education and legal tools. With foundation models’ expanding proliferation in the economy and society, leading developers have agreed to voluntary commitments to ensure their safety, security and trustworthiness. In addition, there is growing regulatory oversight of foundation models in major jurisdictions.

This article will provide an overview of foundation models, and describe how they are regulated under global standards.