OpenAI moves privacy control closer to the customer edge

August 20, 2026

OpenAI introduced an optional Zero Data Retention setting for eligible API customers using its frontier models, under which prompts and responses are not stored after processing, and it previewed Private Safety Processing, a way to check for risky use while keeping customer content on infrastructure the customer controls.

This appears to matter beyond OpenAI because it shifts one of the main objections to cloud AI from trust in the vendor to the design of the deployment. If prompts and responses do not stay with the model provider, and safety checks can run without handing over the content itself, companies handling sensitive material may no longer need to choose as sharply between using frontier models and keeping data under their own control. Through Microsoft's lens, that likely weakens one of Azure's clearest selling points for hosted AI - governance wrapped around access to outside models - because a rival is now making privacy itself part of the API offer rather than only part of the cloud contract.

Microsoft expanded the Microsoft Foundry model catalog by adding DeepSeek-V4-Flash-0731 as a Direct from Azure option and NVIDIA Nemotron 3.5 Lightning through Fireworks and Hugging Face. Microsoft Foundry is the place where Microsoft lines up models from itself and other providers behind Azure billing and controls.

We read this as a sign that the cloud platforms are becoming interchangeable shops for model access faster than the models themselves are becoming interchangeable. For anyone buying generative AI rather than training it, that suggests the practical question keeps moving away from which single vendor has the right model and towards which provider can package outside models with the controls, billing and connections a company already uses.

Google announced Ask Gemini in Google Chat for Workspace, with availability starting on 26 August, to replace the existing Gemini side panel in Chat. Waymo integrated Gemini into its Ojai vehicles for in-car help with cabin settings, nearby places and information about the surroundings. xAI launched Grok 4.6 on Amazon Bedrock with published usage-based pricing in supported AWS regions. Amazon rolled out Alexa+ at no additional cost on compatible Fire TV devices for U.S. customers.