Securing generative AI models on Azure AI Foundry

Microsoft describes its security approach for hosting AI models on Azure AI Foundry and Azure OpenAI Service, clarifying that customer data is not used to train shared models and that models run in isolated, zero-trust environments within customer tenants. The post details pre-release security investigations—including malware analysis, vulnerability assessments, backdoor detection, and model integrity checks—along with guidance to use Microsoft Security tools for ongoing defense and governance.