Zoom’s generative AI/ML models are developed using a diverse and vetted collection of proprietary, synthetic, freely licensed, open source licensed, commercially licensed, and publicly available datasets collected and used beginning in or around 2021 and continuing to the present. Datasets go through a defined legal, compliance, and security review process before use. The training data includes a mix of structured and unstructured information containing a wide range of aggregate data points (ranging from approximately 1 megabyte to 300 terabytes) designed to support generation of high-quality, relevant, and context-aware outputs. Zoom employs internal protocols for data cleaning, processing, and modification to enhance data quality and remove, to the extent possible, personal, IP-protected, or other data not intended for use. The datasets allow Zoom to develop and provide high-quality AI products and features.