Zoom updates its terms of service and a recent blog post to explicitly say that the company won't use “communications-like customer content” to train AI models
Jay Peters / The Verge :
Context & Ripple Effects
Days after language allowing AI training on some customer data drew attention, Zoom narrowed the boundary around communications content in its terms. The change resembles OpenAI’s earlier opt-in position on API data for model training, where customer-controlled data became a distinct policy category.
The wider coverage shows platforms treating AI-training rights as a sensitive terms-of-service issue: Adobe later clarified that customer work would not train Firefly. Zoom’s clarification matters because communications data carries especially direct expectations of confidentiality.
First-order effects
- Zoom customers receive an explicit commitment that communications-like content will not be used to train the company’s AI models.
- Zoom must distinguish communications-like content from other customer data in its AI-training policies and related disclosures.
Second-order effects
- Rival collaboration and cloud software providers face greater pressure to state, in similarly specific terms, whether meeting, chat, and work-product data can train their models.
- Enterprise buyers gain a clearer policy point for procurement and data-governance reviews, making vague references to “customer data” harder to accept.
Third-order effects
- If such clarifications become standard, AI training policies are likely to move from broad platform terms toward data-type-specific consent and exclusion rules.
- The pattern reinforces a public-data permission boundary: firms may increasingly compete on how narrowly and transparently they define permissible inputs for model development.
The trend: AI providers are being pushed to separate access to customer data from permission to use particular categories of that data as model-training input.