iOS 27: Apple’s Siri AI arrives on Google’s engine — Europe and China cannot use it

iOS 27 landed on Monday as the iPhone update Apple has been working toward since it first described a genuinely intelligent Siri in 2023. The new assistant reads emails without a specific prompt, checks the calendar, executes actions across multiple apps in sequence, and holds a conversation through more than twenty consecutive turns without losing context. Those capabilities may sound like modest increments. In practice, no Siri released before this week could do any of them reliably. The gap between what Apple described at keynotes and what Siri actually did when spoken to has been a source of genuine frustration — and persistent ridicule — since 2011. iOS 27 closes that gap. It does so by admitting, quietly but legibly, that closing it required help from a competitor.

Apple signed a multi-year deal with Google reportedly worth around one billion dollars per year to run Siri AI on Google’s Gemini 2.5 Pro. The scale of the arrangement is striking; the reason for it is more instructive. Gemini 2.5 Pro carries roughly eight times the parameter count of Apple’s own in-house foundation models — a gap that says something candid about where Apple honestly assessed its position in the large-language-model race. Apple did not license Gemini because it lacked an AI team. It did so because the models that team has built over the past several years cannot yet produce an assistant that performs at the level Siri AI requires. The deal was announced at WWDC 2026 in June; the September 14 public release translates that announcement into a technical fact. Apple is now running its flagship consumer assistant on its oldest direct competitor’s infrastructure.

User data does not pass to Google in a readable form. Apple routes all queries through its Private Cloud Compute infrastructure — a system designed so that Google processes each request without associating it with the identity of the person who sent it. The architecture uses encrypted enclaves that segregate query processing from anything that could link it to an Apple account or device. Whether that fully satisfies users who prioritize privacy depends on how much trust they extend to systems they cannot personally inspect, but the technical intent is considerably more deliberate than a straight API call to a third-party server. Apple has been publishing the specifications for Private Cloud Compute since its initial announcement, and independent security researchers have begun reviewing them. The commercial communications, characteristically, avoid the hardest follow-up questions.

For a company that designs its own silicon, writes its operating systems from scratch, and has built vertical integration into its competitive story for two decades, the Google arrangement carries a particular weight. Apple has integrated its own Neural Engine into every chip it ships since 2017, trained its own large language models, and positioned Apple Intelligence — the broader AI initiative under which Siri AI sits — as the reason to upgrade to a newer iPhone. The implicit promise running through every Apple Intelligence presentation since WWDC 2023 was that Apple’s on-device approach would eventually make cloud-dependent models look privacy-invasive by comparison. That Apple is now running its marquee AI feature on Google’s cloud model does not make the on-device work irrelevant. It does clarify the distance between what Apple has been promising and what Apple’s AI can currently produce unassisted. The deal was the announcement of June; the shipping product is the reckoning of September.

iOS 27 is a free update for every iPhone still receiving software support. Siri AI is not. The minimum requirement is an iPhone 15 Pro, which ships with the A17 Pro chip — the first consumer iPhone processor Apple judged capable of meeting the ML performance floor the new Siri requires. That gate covers approximately three and a half device generations, or between 150 and 200 million handsets currently in use. Older iPhones receive iOS 27’s other improvements — a redesigned notification system, updated child safety filters, performance gains throughout the operating system — but the assistant on those devices stays as it was. Accessing Siri AI on a compatible device requires joining a waitlist at launch, an additional step beyond hardware eligibility that Apple has not fully explained. The iPhone 17 lineup, announced alongside iOS 27 and going on sale October 23, ships with a larger on-device model that Apple says produces faster and more capable Siri AI responses than even the A17 Pro hardware can deliver. The hardware gate and waitlist requirement together concentrate the most visible AI improvements at the recent end of Apple’s installed base, which is not accidental.

With the necessary permissions granted, the new Siri connects to email and calendar without requiring a specific command structure. You can ask it to find a message from last week about a dentist appointment and reschedule it, and it will attempt to handle that across two separate apps in a single request rather than directing you to open each application yourself. It can draft email replies from a verbal summary, add items to reminders mid-conversation, and work across supported scheduling and booking apps on compound requests. On macOS 27, Siri AI integrates into Spotlight, effectively turning the system-wide search bar into a task executor: type or speak a compound request and receive an action rather than a list of results. Apple states that the assistant sustains coherent dialogue across more than twenty consecutive turns — substantially longer than any prior Siri held a thread before losing context. The multi-app, multi-step capability is what the original 2011 Siri demos implied was coming; iOS 27 is when it actually arrived.

The iOS 27 launch version of Siri AI is English-only. French, Japanese, Korean, Portuguese, and Spanish are scheduled to follow in October. Mandarin Chinese has no confirmed date. That language map tells you what Monday’s release actually covers: it is primarily a US and English-speaking deployment, with the rest of Apple’s major markets following in waves Apple has only partially committed to. A feature reaching between 150 and 200 million eligible devices that operates in one language on launch day is both a significant technical undertaking and a significant limitation that Apple’s presentations tend to minimize. The pacing is consistent with how earlier Apple Intelligence features rolled out through iOS 17 and 18, language by language over eighteen months. For a company whose iPhone business is genuinely global, the gap between “Siri AI launched” and “Siri AI launched in your language” is a gap hundreds of millions of users will feel for months — and, in some markets, possibly into 2027.

The most consequential absence in the iOS 27 Siri AI launch is geographic. Apple will not activate Siri AI on any iPhone or iPad in the European Union, and has given no timeline for when it will. China is also excluded, for separate regulatory reasons Apple has not detailed publicly. The EU situation stems from an unresolved dispute over the Digital Markets Act. Under the DMA, Apple is designated a gatekeeper platform and required to open the device-level hooks that allow Siri to read messages, initiate purchases, access files, and execute actions across third-party apps — to any competing AI assistant that requests equivalent access on the same terms. Apple argued it could not grant that access safely without a structured intermediary layer, and designed a framework it called Trusted System Agent: a proposed secure bridge intended to let rival AI systems reach the phone’s sensitive functions without having raw access to the device’s underlying data streams. The European Commission reviewed the proposal and declined it. Apple then chose to ship without the EU rather than redesign its regulatory position to meet the Commission’s requirements.

Mac users in Europe are not excluded. macOS 27 falls under different DMA enforcement rules, and European Mac owners gain Siri AI in Spotlight from the September 14 release alongside users everywhere else. That carve-out makes the iPhone exclusion read less like a genuine technical constraint and more like what it likely is: a negotiating position. Apple has the privacy infrastructure, the model, and the hardware deployed and selling throughout Europe. What it does not have, for iPhones and iPads in the EU, is a regulatory arrangement it considers acceptable. The European Commission has not indicated publicly what, if anything, would lead it to revisit its rejection of the Trusted System Agent proposal. Apple has been consistent in saying it cannot safely comply with what the DMA as currently interpreted requires. The standoff has no visible resolution date, and it has been running since the DMA came into full enforcement in 2024.

There is a specific irony in the iPhone’s core AI running on Google’s infrastructure. Apple and Google have competed directly in mobile operating systems since 2007. Google Assistant and, more recently, Google Gemini — the very model under Siri AI — compete directly with Siri in the AI assistant market that Apple created with the original 2011 launch. The two companies also maintain a long-standing separate commercial deal: Google has paid Apple to remain the default search engine in Safari, an arrangement worth an estimated twenty billion dollars a year at its recent peak and currently under antitrust review by the US Department of Justice. The Siri AI deal is, in structural terms, a parallel arrangement: Apple routes a competitor’s AI capability through its own user-experience layer, presents it under Apple’s branding, and the underlying model belongs to someone else. The difference is visibility. When Google powers the Safari search index, most users do not experience it directly. When Siri AI answers your question, the interface is entirely Apple’s — but the model underneath is Google’s Gemini.

Siri AI on Gemini 2.5 Pro places the iPhone back in a credible position in the consumer AI assistant market. The earlier ChatGPT integration on iOS — announced at WWDC 2024 — gave users an opt-in path to OpenAI‘s models, but it sat alongside Siri rather than replacing it. Siri AI on Gemini is a deeper integration: it sits inside the default assistant and has access to the device’s core permissions layer. Samsung’s Galaxy AI on Android has been built around Google’s models for the past two years; the iPhone is now using the same underlying model stack as its most direct hardware competitor, while presenting it through a different experience layer. Microsoft Copilot on Windows has moved in a similar direction across the same period. The consumer AI assistant landscape in late 2026 converges around a small number of frontier models — primarily those from Google, OpenAI, and Anthropic — with hardware makers competing on how they present those models to users. Apple’s answer, with iOS 27, is that the presentation is the product: tighter integration, stricter privacy architecture, and a surface so polished that the provenance of the underlying model becomes secondary to the experience of using it.

iOS 27 is available as a free download. The first real test of whether Siri AI changes how people actually use their iPhones will happen in the markets where Apple can turn it on — currently most of the world, with the European Union and China excluded. Whether the EU standoff generates enough pressure on Apple to revise the Trusted System Agent proposal, or on the European Commission to revisit its requirements, is likely to shape the regulatory conversation around AI and platform access through most of next year. For now, Apple has shipped the most capable Siri in the iPhone’s fifteen-year history. The model running underneath it is Google’s. That fact does not diminish the engineering work that surrounds it — but it does give an honest account of where Apple’s AI stands, and how far it still has to go on its own.

Tags: Apple, Raphael, smartphones, Siri, Digital Markets Act, iOS 27