product updateApple

Apple Confirms Siri's New AI Features in China Run on Alibaba's Qwen Models, Not Gemini

TL;DR

Apple has confirmed that its new Siri AI features in China will be powered by Alibaba's Qwen models rather than Google Gemini, which handles Siri AI elsewhere. A published guide indicates users will need separate Qwen accounts, suggesting the privacy protections built into Apple's Private Cloud Compute won't apply.

2 min read
0

Apple has confirmed that Siri's new AI features in China will run on Alibaba's Qwen models rather than Google Gemini, which powers Siri AI in most other markets. The disclosure, reported by Reuters and first spotted in an updated Apple support guide, raises questions about whether Chinese users will receive the same privacy protections as users elsewhere.

In most countries, Apple's Siri AI features are powered by Google's Gemini models, running either on Apple's own Private Cloud Compute (PCC) infrastructure or on dedicated Google servers built to PCC-equivalent privacy standards. Under that arrangement, Apple states that neither it nor Google can access user interactions with Siri.

China does not permit this setup. Chinese law requires that AI services offered in the country use domestically approved models, a rule widely understood to give the government visibility into AI usage that foreign-hosted, encrypted systems would prevent. Apple has therefore partnered with Alibaba, integrating the company's Qwen models into Siri and Apple's Writing Tools for Chinese users.

According to Apple's updated Chinese-language support guide, users who opt in can use Qwen through Siri for more detailed responses to certain requests, including analysis of photos and documents. Writing Tools will also be able to call on Qwen to generate text or images from a text description. Reuters reports the guide is specific to Mac, but the same Qwen integration is expected to extend to iPhone, iPad, and Vision Pro in China.

A detail in the guide points to a meaningfully different privacy model: Mac users must have their own separate Qwen account and be signed into it to use the feature. This stands in contrast to the Gemini-powered version elsewhere, where Apple states user interactions remain inaccessible to both Apple and Google. Requiring a standalone third-party account suggests Alibaba, and potentially Chinese regulators, may have visibility into queries that Apple's PCC architecture is specifically designed to prevent in other markets.

Apple has not detailed what data protections, if any, apply to the Qwen integration, and the company has not commented beyond the support guide itself. Alibaba has not issued a statement on the arrangement's privacy terms.

What this means

This is not a new model release — it's a regional deployment decision forced by Chinese law, which bars foreign AI models from operating domestically. Apple's dependence on Alibaba's Qwen underscores a structural reality for any global tech company operating in China: local infrastructure and licensing requirements often mean privacy architecture cannot be uniform across markets. The requirement for a separate Qwen account is the clearest signal that Apple's zero-access privacy guarantee, central to its marketing of PCC and Gemini-powered Siri elsewhere, does not extend to Chinese users. Expect scrutiny from privacy researchers and possibly regulators outside China over whether Apple's global privacy claims about Siri AI need clearer country-specific caveats.

Related Articles

product update

Perplexity Launches Hybrid Compute for Mac, Splitting AI Tasks Between Cloud and Local Models

Perplexity's Mac app now supports Hybrid Compute, which starts tasks in the cloud and shifts sensitive steps to a local model running on-device. The feature requires Apple silicon with at least 24GB of unified memory and uses an open-sourced on-device PII classifier to mask private data before any cloud processing.

product update

AWS Publishes Reference Architecture for Multimodal WhatsApp Ordering Agents Using Bedrock AgentCore and Nova 2

AWS published a reference architecture showing how to deploy a WhatsApp ordering assistant on Amazon Bedrock AgentCore, using Nova 2 Lite for text and Nova 2 Sonic for voice, with shared cross-channel memory and MCP-based tool access to backend systems.

product update

Gemini Overlay on Android Adds Minimize Button for Multitasking Bubble

Google is widely rolling out a new Minimize button for the Gemini overlay on Android, which collapses conversations into a floating bubble users can drag or tap to expand. The feature currently supports six fixed positions and has a reported bug that resets bubble placement after each minimization.

product update

OpenAI Lists GPT-6 Astra Pro on OpenRouter: Same Model, Higher-Compute Reasoning Mode

GPT-6 Astra Pro, now listed on OpenRouter, is the existing GPT-6 Astra model configured to run with reasoning.mode set to 'pro' for higher-quality output on complex tasks. It carries a 1M-token context window and tiered pricing from $5/$25 to $20/$100 per million input/output tokens depending on the serving tier.

Comments

Loading...