improvedhigh materialitytrainingAug 3, 2026 to Aug 28, 2026
model training improved from high/training permitted to low/no training claim.
Stance key: model training. AIRIN emits this only when before and after stance events both carry citations.
Before · training permitted
high
“(h) use the Service to develop, train, or improve a competing product, service, or AI model, or to assist a third party in doing so;”
§ 5 (Use Restrictions)
Open citationAfter · no training claim
low
“Provider Training on customer data Default retention Configuration we use OpenAI (direct API) Not used to train OpenAI models. See OpenAI API data controls . Up to 30 days for safety and abuse monitoring under the standard API. Where eligible and approved, we use OpenAI's Zero Data Retention (ZDR) on supported endpoints. Anthropic (direct API) "Anthropic may not train models on Customer Content from Services." See Anthropic Commercial Terms and Anthropic DPA . Trust-and-safety logs may be retained for a limited period. We use Anthropic's commercial API under their Commercial Terms and DPA. Microsoft Azure OpenAI Service Not used to train OpenAI's or Microsoft's models. See Data, privacy, and security for Azure Direct Models in Microsoft Foundry . Up to 30 days of abuse-monitoring retention by default, with content logging. Where we are eligible, we use Microsoft's Modified Abuse Monitoring to disable content logging (verifiable as ContentLogging=FALSE ). Microsoft Azure for Anthropic models (Microsoft Foundry) "By default, Anthropic will not use your inputs or outputs from commercial products to train their models." See Data, privacy, and security for use of Anthropic Claude models in Microsoft Foundry . For these deployments, Anthropic (not Microsoft) is the processor. Per Anthropic's terms; Anthropic Claude on Microsoft Foundry is not covered by the Azure OpenAI Zero Data Retention program.”
§ 4.2 (What Model Providers do with that data)
Open citation