iOS 27 opens to third-party AI models

Apple’s AI problem has never really been hardware. The company already owns the devices, the operating systems, and the default apps people use all day. The problem is model quality — Apple’s own generative tools have lagged, while OpenAI, Google, and Anthropic keep moving faster. Now Apple looks ready to change the structure of the whole thing by letting outside AI models plug much deeper into iOS 27, iPadOS 27, and macOS 27. (bloomberg.com) ### What changed today? The new report says Apple plans to let users choose rival AI models to power built-in features across its software, not just occasional chatbot handoffs. That means text generation, image generation, and editing tools inside Apple’s own interfaces could run on third-party models selected by the user. The change is slated for iOS 27, iPadOS 27, and macOS 27 later this fall. (bloomberg.com) ### Isn’t Apple already using ChatGPT? Yes — but in a much narrower way. Apple already has a partnership that lets Siri and some Apple Intelligence features pass a request to ChatGPT when Apple’s own system wants outside help. The new move is bigger because it turns that one-off partnership into a broader architecture where multiple providers can sit underneath Apple’s system features. (macrumors.com) ### What’s this “Extensions” idea? Basically, Apple seems to be building a compatibility layer — reportedly called “Extensions” — that lets external models hook into Apple Intelligence features across the OS. Instead of one baked-in model path, users could pick a provider in Settings, and developers or model companies wou(macrumors.com) feel like a platform feature, not a special exception. (9to5mac.com) ### Which models are in the frame? The names that keep coming up are Google Gemini and Anthropic Claude, alongside OpenAI’s ChatGPT. One report also says Apple has already signed a deal with Google around Gemini-based features for iOS 27, though that part is still in the rumor bucket. The broader point is clearer than any one partner — Apple wants several model options available inside the same operating system. (macrumors.com) ### Why would Apple do this? Because the fastest way to make the iPhone feel competitive in AI is to stop insisting every important model come from Apple. Apple still controls the device, the UI, the privacy framing, and the entry points. But it can outsource the hardest part — frontier model performance — to companies al(macrumors.com)sion of it. (techcrunch.com) ### What does Apple still control? A lot. Apple would still own the operating systems, the default experiences, the permission model, the settings surface, and the hardware integration. Think of it less like Apple giving up the steering wheel and more like Apple turning the engine bay into a slot where (techcrunch.com)reported feature is described across Apple’s own software layers. (bloomberg.com) ### What’s the catch? Opening the stack creates new headaches — privacy disclosures, billing, moderation differences, latency, and the weirdness of one phone giving different answers depending on which model the user picked. It also risks making Apple’s own models look second-r(bloomberg.com)kup option. (bloomberg.com) ### So why does this matter? If this ships as described, Apple Intelligence stops being a single assistant strategy and becomes a distribution layer for the best models Apple can attract. That would be a real policy shift for a company that usually treats core system capabilities as something it builds, names, and controls end to end. In AI, turns out, control may matter less than relevance. (bloomberg.com)