Apple SVP Craig Federighi clarified the company's Google partnership during a post-WWDC tech talk, explaining that Siri AI in iOS 27 doesn't rely on Google's Gemini models or infrastructure. Federighi emphasized Apple builds its own Assistant experience deeply integrated across iOS, iPadOS, and macOS, with unique privacy-focused architecture.

The system uses on-device models for speech recognition and voice synthesis, plus a System Orchestrator that coordinates requests through the App Toolbox and Spotlight Semantic Index. For complex queries, Apple's Private Cloud Compute extends device privacy to cloud processing, ensuring requests aren't stored or accessible to Apple or anyone else.

Apple's third-generation Foundation Models include AFM Core, AFM Core Advanced, AFM Cloud, AFM Cloud Image, and AFM Cloud Pro. These custom-built models for Apple Silicon were trained on proprietary data and refined from Gemini frontier models, though the Pro variant uses NVIDIA GPUs in Google's cloud while maintaining Apple's privacy guarantees.

For current events and world knowledge, Apple taps its own World Knowledge Service rather than Google Search. This hybrid approach lets Apple match user requests to the optimal model while preserving its privacy-first positioning in the competitive AI assistant market.