What the Siri AI Redesign Really Is
Apple’s Siri AI redesign is a ground-up rebuild of the assistant as a deeply integrated, on-device and cloud-hybrid system that uses personal context intelligence, custom Apple Foundation Models, and privacy-first infrastructure to become a primary on-device AI assistant rather than a thin wrapper over external chatbots. At Apple WWDC 2026, this new Siri arrived with its own dedicated app and system-wide hooks across iOS 27, signaling a shift from “voice shortcut” to core interface. Craig Federighi stressed that the assistant is not Gemini in disguise, even though Apple collaborates with Google on model training. Instead, Siri runs on Apple’s own models and infrastructure, supported by a system orchestrator that routes tasks between device and cloud. The aim is a Siri that understands your device, your content, and your habits, and can respond in natural language without sending every request to a remote server or a separate chatbot app.
How Apple Uses Gemini Without Becoming Gemini
Many assumed the Siri AI redesign meant Gemini inside a new skin, but Apple describes a more nuanced relationship. Federighi told reporters that Apple does not ship the Gemini app, client code, or Google’s deployment stack on iOS. Instead, the company built a family of Apple Foundation Models (AFM) that sit on-device and in the cloud. These models were refined using outputs from Gemini frontier systems, but Siri does not call Gemini Flash, Gemini Pro, or Google’s image models at runtime. According to PCMag, Amar Subramanya explained that reinforcement learning and Gemini outputs served as a kind of training mentor, not a permanent backend. Apple’s own world knowledge service, not Google Search, answers up-to-date questions. Gemini’s role is closer to that of a teacher in the lab than a live engine in the product, which helps explain why Siri can feel different from a generic chatbot while inheriting some of its intelligence.
Personal Context Intelligence: Siri as a True System Assistant
The heart of Apple’s strategy is personal context intelligence: using what your device knows about you to give relevant, precise responses. Siri now lives as a dedicated app and a system assistant that plugs into messages, documents, photos, and settings, all coordinated by a system orchestrator. This orchestrator inspects each request, decides whether AFM Core on-device can handle it, and only escalates to cloud models when needed. Because Siri sits inside the OS, it can act on your behalf rather than only chat. It can interpret emails alongside calendar events, or match voice requests with on-device files, without exposing that data to external services. Apple’s redesign tries to make Siri feel less like a bridge to another AI and more like a native agent that understands your environment, understands what you mean by “that photo from last week,” and responds in seconds without constant round trips to third-party servers.
Hybrid On-Device and Private Cloud Compute
Behind the scenes, the Siri AI redesign is anchored in a hybrid architecture that balances on-device AI assistant features with cloud scale. Apple’s AFM Core and AFM Core Advanced models run directly on Apple silicon for quick tasks, multimodal dictation, and the assistant’s new expressive voice. When the system orchestrator detects tougher jobs—complex workflows or image editing—it routes them to AFM Cloud, AFM Cloud Image, or AFM Cloud Pro. Those cloud models run in Apple’s Private Cloud Compute (PCC), which Federighi describes as an extension of the iPhone’s privacy guarantees. Requests are processed on servers that Apple says even its engineers cannot inspect. PCMag reports that AFM Cloud Pro depends on Nvidia GPUs inside Google’s infrastructure, but PCC fences off user data. The effect is a tiered system where everyday Siri actions stay local, while heavier tasks borrow massive compute power without handing your personal information to external AI platforms.
