Microsoft shuts an older RAG path as local AI reaches public service delivery

October 15, 2026

Microsoft retired Azure OpenAI On Your Data and told customers to move those workloads to Microsoft Foundry Agent Service with Foundry IQ. Retrieval-augmented generation means a model answers with material pulled from a company’s own documents instead of only from what it learned in training. Foundry is Microsoft’s model and agent platform, and Foundry IQ is the part that fetches and prepares that material for an answer. This appears to matter beyond one product rename because customers that built around a simple add-on now have to migrate into a fuller agent stack with Microsoft’s control layer around it. It also suggests that governed enterprise AI is being sold less as access to one model feature and more as a managed system for retrieval, orchestration and policy.

Google DeepMind and Singapore’s GovTech launched on-device scam detection with lightweight Gemma models running locally, keeping user data on the device while monitoring conversational patterns and blocking scams. On-device here means the model runs on the user’s own hardware rather than sending each interaction to a cloud service. This is likely to change what buyers compare: local execution is no longer only a privacy extra, but a way to ship real-time public-facing AI with lower data exposure and tighter control over where information stays. For Microsoft, that sharpens the pressure to offer local model access through ordinary software surfaces, not only through cloud administration.

Anthropic opened an office in Singapore and signed a memorandum with the government there on AI safety, security, model evaluation and responsible adoption. This looks less like a market-entry footnote than another sign that frontier model providers now need state relationships as part of access, oversight and deployment, especially in places that want adoption and supervision together.

Anthropic published stronger election safeguards for Claude ahead of the US midterms. Alibaba Cloud introduced QwenWork for Chinese e-commerce merchants. Alibaba Cloud also announced Cloud Use to run AI agents as governed cloud workloads. Barclays downgraded Dell, HPE and NetApp to hold, citing hardware revenue driven by memory pricing rather than broad AI spending. US stocks fell after inflation data, with CoreWeave, Oracle, Micron, AMD and Intel among the decliners. xAI launched Grok Voice 3.0. Graham "Graham Dot Zip" showed an AI job candidate agent tangling with an AI interview bot.