Apple’s new AI split means your iOS app could behave differently in China
The New Stack Amanda Caswell ● Covered by 2 sources
Apple is making a China-only AI setup for iPhones and Macs. That could make the same app answer differently depending on where it's used.
Based on reporting by The New Stack, Amanda Caswell — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Apple is no longer treating its AI stack as one clean global system. In China, Reuters says the company has built a separate model with help from Alibaba, because the usual mix of outside AI services simply isn’t available there.
The twist is that Apple isn’t just plugging in a partner’s model and calling it a day. Reuters says the China version is expected to work with Alibaba’s Qwen and possibly Baidu technology, while Apple still trains its own regional model. Apple and its Chinese partners haven’t said which model will do what inside Apple Intelligence, and that silence matters.
For most of the world, Apple handles AI requests on device or on its private servers using its own models, with Google’s Gemini handling more complex cloud requests through a multi-year licensing deal formalized in early 2026. Siri users can also tap ChatGPT through an optional extension. None of that setup carries over to China, where ChatGPT, Gemini and Anthropic’s Claude can’t offer the same services.
Apple has already left one clue about how this might work. Reuters reported on August 8 that the company had posted a Chinese-language guide showing Mac users how to connect Qwen to Siri and Writing Tools, then removed the page without explaining why. Baidu technology is also said to be part of the planned system.
For developers, the problem is ugly in the boring way that causes the most pain: the same iOS app could ship, test and pass on one device, then behave differently in China without any code change. If Apple’s Foundation Models framework points to a different system model there, the same prompt could yield a different answer, a refusal, or a different structured output. Apple’s registration with the Cyberspace Administration of China in mid-July cleared a major hurdle, but it also adds a compliance layer that can filter out more than just obvious red flags. Without clear documentation, the blame could fall on Apple, Qwen, Baidu, or the app itself.
My take — AI-written commentary, not fact-checked reporting
This is the part of the AI story people keep trying to file under “regional customization,” which is a neat way to make a regulatory headache sound civilized. It’s really a reminder that open, portable app behavior gets messy fast once model choice becomes policy choice. Developers don’t need another mystery box; they need to know which box is doing the refusing.
Read more about this at: The New Stack