What the New Apple Siri AI Is—and Is Not
The new Apple Siri AI is Apple’s rebuilt assistant system that combines on-device intelligence, private cloud models, and deep OS integration to provide context-aware help without turning into a generic chatbot. It is designed to answer questions, control apps, and perform tasks while keeping personal data local whenever possible. At WWDC 2026, Apple framed this overhaul as part of its broader Apple Intelligence push, which many observers assumed meant Siri would become a front end for Google’s Gemini models. Craig Federighi pushed back on that idea in a post-keynote briefing, stressing that “we use none of the models that Google deploys to their customers, nor do we use the infrastructure and means by which they deploy models to their customers.” In other words, Gemini helped train Apple’s models, but it is not running your requests.
Inside Siri’s On‑Device Intelligence and System Orchestrator
At the heart of the WWDC 2026 Siri redesign is a layered family of Apple Foundation Models that stretch from phone to cloud. AFM Core and AFM Core Advanced sit on-device to handle everyday language, dictation, and Siri’s new more expressive voice. For harder requests, the “system orchestrator” decides whether to escalate to AFM Cloud, AFM Cloud Image, or AFM Cloud Pro. This component weighs task complexity and privacy needs, then routes the request either to the phone or to Apple’s Private Cloud Compute servers. When a request does go to the cloud, Apple says data is isolated so even its own engineers cannot see it. The result is an assistant that feels faster for local actions and more capable for complex tasks, while keeping the bulk of personal context, like messages and documents, on the device rather than in a remote AI stack.
How Gemini Trains Siri Without Powering It
Apple’s collaboration with Google happens behind the scenes rather than in the Siri client you speak to. Amar Subramanya, Apple’s vice president of AI engineering and a former Gemini lead, explained that Apple’s AFM family is “custom builds for Apple Silicon, trained using proprietary data, and refined using outputs from Gemini frontier models.” In practice, that means Siri’s models learned from feedback and examples produced by Gemini during training, through reinforcement learning that rewards better answers and penalizes weak ones. PCMag reports that Google’s cloud, equipped with Nvidia GPUs, also helped scale Apple’s Private Cloud Compute for the demanding AFM Cloud Pro model. But Federighi was explicit that the shipping assistant does not call Gemini APIs or Google Search. Gemini is more like a teacher from Siri’s past, not a hidden engine running every time you say “Hey Siri.”
From Secondary Chatbot to Core OS Feature
This redesign is Apple’s response to years of complaints that Siri lagged behind Google Assistant and tools like ChatGPT. The earlier Siri–ChatGPT tie-in often made Siri feel like a pass-through, repeatedly asking whether it should “use ChatGPT to answer that.” With the new Apple Siri AI, Apple is trying to erase that middleman feeling and position Siri as a core part of the operating system. Instead of living in a separate app, Siri is wired into iOS through the system orchestrator and direct access to personal context such as apps, content, and settings. Apple is betting that tight integration and on-device intelligence, backed by Private Cloud Compute when needed, can outshine the raw breadth of external chatbots. The goal is an assistant that understands your device and your habits better than competitors, while still keeping Apple’s privacy promise front and center.
