iOS 27 Launches With Siri from Google Gemini, EU Blocked

A fifteen-year overhaul delivered in haste

Apple began distributing iOS 27 at 10 a.m. Pacific time on Monday, September 14, featuring Siri AI rewritten from the ground up and a strategic decision the company refrained from detailing during the event: the assistant is a distillation of knowledge from Google Gemini. Google's models have trained Apple's on-device models, although Google’s servers are not consulted in runtime for most requests. For a company that spent a decade building differentiation based on privacy and vertical integration, outsourcing the backbone of its assistant marks Cupertino’s most public concession that it has lost the generative cycle. This is the first structural overhaul of Siri since 2011.

What changes inside the iPhone

The Siri AI is conversational, supports multiple turns, maintains session context, and cross-references data from Mail, Messages, and Photos to answer personal questions. Apple has also opened hooks for the assistant to perform actions within apps, which was the weakest point of the previous product. The release has three critical caveats for companies operating fleets: the feature is in beta and requires users to join a waitlist, support is limited to the iPhone 15 Pro—three and a half generations into the lineup—and only English is available at launch. A CIO responsible for thousands of devices will spend weeks unable to predict how many employees have actually joined the queue.

Geographic restrictions and corporate implications

The Siri AI is unavailable for iPhones in the European Union and China. In Brussels, the reason is the Digital Markets Act, which Apple has already used as justification for delaying AI features in the continent. In Beijing, the issue is regulatory and commercial: a public reliance on Google’s technology does not pass muster under the Cyberspace Administration of China, and Apple is still trying to finalize a local deal with Baidu or Alibaba. For a multinational with teams in Munich, Frankfurt, and Shanghai, this fragmentation compels a reevaluation of device policies by country: employees of the same company will have iPhones with different functional capabilities. In Brazil, the release arrives in English and without a visible local waitlist, which practically means the Portuguese version will be delayed by months.

What the Gemini distillation really delivers

Distillation is a recognized technique: a large model trains a smaller model to mimic its behavior, which reduces cost and latency without requiring the smaller one to undergo pre-training. It works well for repetitive tasks but loses accuracy on rare queries. For the user requesting "summarize the board email," the drop is imperceptible. For the user asking, "compare the terms of this contract to the standard we approved in July," the decline could mean the difference between a useful response and a hallucination. It’s no surprise that Apple has reserved complex queries for the Private Cloud Compute, which runs larger models on its own servers.

What remains for competitors

Today's announcement puts pressure on Samsung and Google to respond in the next October launch window. The Galaxy S26 already runs Gemini natively, without a waitlist and geographic block, and Samsung is expected to leverage this as a bargaining chip in renewals with banks and insurers purchasing Android fleets. Anthropic and OpenAI, who have spent the year trying to secure on-device contracts, emerge weakened: the most strategic round of the cycle, Siri, has gone to Google. If Apple wishes to regain differentiation, the next move is likely to be a proprietary model trained on internal ASICs. The cycle of iOS 28 will reveal whether this correction takes two years or five.