Bill > HB823

This bill requires developers of generative artificial intelligence (AI) systems to publish detailed documentation about their training data on their websites by January 1, 2026, and before each subsequent substantial modification. The bill defines generative AI as artificial intelligence that can create synthetic content like text, images, video, and audio that mimics the characteristics of its training data. The documentation must comprehensively disclose details about the training datasets, including the sources or owners of the data, how the data supports the AI system's purpose, the number of data points, label types, data domain status (public or protected), data acquisition method, personal information inclusion, data processing methods, collection time period, first use date in development, and whether synthetic data generation was employed. The requirements apply to generative AI systems released for public use since January 1, 2022, with specific exemptions for systems exclusively used in contexts like hospital medical staff operations, physical safety, confidential information protection, fraud detection, national security, and military purposes. By mandating this transparency, the bill aims to provide insight into the origins and composition of training data used in generative AI systems, potentially helping users understand the AI's potential biases, limitations, and development process.