Apple reveals new AI architecture built around Google Gemini models

Most of the energy went into privacy and EU access rules. Apple’s delay in the EU was widely read as a fight over whether Siri-grade access to messages, app state, and cross-app actions must also be available to third-party assistants under the Digital Markets Act. People who buy Apple’s framing see a real security problem in giving any assistant broad read and act permissions across the phone. People who do not buy it see classic Apple lock-in dressed up as privacy. What the thread largely agreed on is the concrete tradeoff: this is not just another chatbot picker. The useful version of this product needs deep OS hooks, and once those hooks exist the hard question is who gets them and under what consent model. The other strong theme was that Apple is behaving like Apple. It is late on frontier models, but not necessarily late on product strategy. Several commenters argued the model itself is becoming a commodity for mainstream assistant tasks, while the durable value is the harness around it: app integration, routing, latency, permissions, and whether users feel safe letting it touch their personal data. That is why Google can be both a competitor and the likely supplier. Google has the small multimodal models, the infrastructure, and the willingness to let Apple run a privacy-wrapped version at Apple scale. Skeptics still think this undercuts Apple’s branding and may leave users with Gemini’s familiar problems like hallucinations and verbosity. But the thread’s center of gravity was that Apple is betting people will judge the assistant by the system experience, not by whose weights sit at the bottom.