Apple lets users pick AI models
- Apple is reportedly preparing iOS 27, iPadOS 27, and macOS 27 to let users choose outside AI models for Apple Intelligence tasks. - The new system is internally called “Extensions,” and reports say it could cover Siri, Writing Tools, and image features this fall. - That would push Apple toward being an AI router and trust layer, not just a builder of one in-house assistant.
Apple’s AI story may be getting a lot less Apple-only. The new twist is simple but important — instead of forcing one assistant or one model everywhere, Apple is reportedly building a way for users to pick which outside AI service handles parts of Apple Intelligence. If that ships in iOS 27, iPadOS 27, and macOS 27 this fall, the company stops looking like a single-model AI vendor and starts looking more like a platform that brokers access to several models. (bloomberg.com) ### What changed? The reported change is a new internal feature called “Extensions.” The idea is that Apple Intelligence tasks could be routed through a user-selected provider in Settings instead of defaulting to Apple’s own models plus the narrower ChatGPT handoff Apple already has today. Reports point to text generation, editing, image work, and Siri responses as the obvious use cases. (bloomberg.com) ### Isn’t Apple already using ChatGPT? Yes — but only in a limited, permission-based way. Apple announced in 2024 that Siri could hand certain requests to ChatGPT, and that Writing Tools could use it too, with the user asked before data is shared. (bloomberg.com)ltiple Apple Intelligence features rather than a one-off escape hatch. (cnbc.com) ### Which models are in the frame? The names that keep coming up are Google’s Gemini and Anthropic’s Claude, alongside ChatGPT. Several reports say Apple has tested or discussed integrations with Google and Anthropic, and one says Apple has already signed a Google deal tied to Gemini-based features. The catch is that no(cnbc.com)e before WWDC or the fall release. (macrumors.com) ### Why would Apple do this? Because Apple is behind in frontier models, but ahead in distribution. It has the installed base, the operating systems, the hardware, and the trust pitch. So the practical move is not necessarily to beat OpenAI, Google, and Anthropic model-for-model. It’s to mak(macrumors.com)t is basically the App Store logic applied to generative AI. (techcrunch.com) ### What happens to Apple’s privacy pitch? That part probably stays central. Apple’s existing Apple Intelligence system splits work between on-device models and Private Cloud Compute for heavier requests, with the company framing PCC as cloud processing that preserves the (techcrunch.com)trust boundary legible — what stays local, what goes to Apple’s cloud, and what leaves for OpenAI, Google, or Anthropic. (security.apple.com) ### Why does this matter for Siri? Because Siri’s biggest problem has not been branding. It has been capability. Apple spent 2024 and 2025 promising a more useful, context-aware Siri, then struggled to deliver on the timeline people expected. Letting users swap in stronger outside models is a shortcut around that bo(security.apple.com)ot best-in-class yet. (9to5mac.com) ### What’s the real shift here? The real shift is architectural. If this lands, Apple Intelligence becomes less like one assistant and more like a control plane — a layer that decides which model should do what, under Apple’s UI, permissions, and privacy rules. That could (9to5mac.com). (bloomberg.com) ### Bottom line Apple seems to be conceding something important — users may care less about whose model they’re using than whether the feature works. If that’s right, giving people a choice is not a retreat. It’s a very Apple way to stay in the middle of the AI stack.
