← Back to TITS Homepage

In what will be remembered as one of the most consequential product announcements in Apple's history, Tim Cook took the stage at Apple Park this morning for his final WWDC keynote as Chief Executive Officer. The event delivered exactly the AI reset that analysts had been demanding for two years: a completely rebuilt Siri running on a custom 1.2-trillion-parameter version of Google's Gemini model, a new multi-AI Extensions framework that lets users set Claude, ChatGPT, or other models as their default assistant, and a sweeping OS rebrand to iOS 27 across all Apple platforms. Cook will transition to Executive Chairman on September 1, handing the CEO title to Hardware SVP John Ternus — making today's keynote both a product milestone and a passing of the torch.

The announcement ends nearly two years of public embarrassment for Apple in the AI race. After a $250 million settlement over delayed Apple Intelligence features and repeated failures to ship a conversational Siri that could compete with ChatGPT and Claude, Apple concluded that building frontier AI models in-house was not a battle it could win quickly enough. The solution it landed on is audacious in its pragmatism: license the best model available, wrap it in Apple's design and privacy infrastructure, and ship.

The New Siri: Gemini Under the Hood

The rebuilt Siri is powered by what Apple is calling a "custom deployment" of Google's Gemini model — a version reportedly fine-tuned for Apple's on-device context, privacy constraints, and Apple ecosystem integrations. Google's models serve as the cloud reasoning backbone, while Apple's own silicon continues to handle on-device tasks like real-time transcription, sensitive data processing, and offline queries. The commercial terms are substantial: Apple is paying Google approximately $1 billion per year under a multi-year agreement, a figure that underscores how seriously Apple views the deal and how urgently it needed to close the AI gap.

Functionally, the new Siri bears little resemblance to its predecessor. Users now interact with it through a dedicated Siri app that features a full chat interface — text, voice, and image input — directly analogous to the ChatGPT and Claude interfaces that have defined AI consumer products since 2023. A new system-wide "Search or Ask" gesture triggers Siri from any app or screen, and Dynamic Island integration on iPhone 16 and newer models provides ambient status indicators while Siri is processing longer requests. The assistant now handles multi-step compound commands — asking it to "summarize my last three emails from the Toronto office, draft a reply to the most urgent one, and add a follow-up reminder for next Tuesday" works as a single instruction.

Key Insight: Apple is paying Google ~$1 billion per year to power Siri with Gemini — a frank acknowledgment that building a frontier AI model from scratch was not Apple's competitive path, and that speed to market now outweighs first-party purity.

AI Extensions: A Multi-Model Framework

Perhaps the most structurally significant announcement for the broader AI industry is iOS 27's new AI Extensions framework. Apple is opening a formal API that allows users to designate any participating AI model as their system-level default assistant. At launch, three models are supported alongside Siri itself: Google Gemini, OpenAI's ChatGPT, and Anthropic's Claude — marking the first time Claude has been available as a system-level assistant on iPhone.

This is a meaningful competitive development for all three AI labs. Distribution through iOS, which ships on over a billion active devices, dwarfs the reach of any standalone app. The Extensions framework also gives users the ability to invoke specific models for specific contexts — using Claude for longer analytical tasks, Gemini for Google Workspace integrations, and Siri for device-native actions — creating a genuine multi-model workflow layer within the operating system rather than forcing a single assistant for all use cases.

iOS 27 and the Liquid Glass Design Language

Apple simultaneously announced that all of its operating systems are moving to version 27 — iOS 27, iPadOS 27, macOS 27, watchOS 27, tvOS 27, and visionOS 27. The version number unification is intentional: Apple wants the software stack to feel like a single continuous platform rather than a collection of parallel OS lines. iOS 27 Beta 1 was seeded to developers the same afternoon as the keynote.

The visual language underlying iOS 27 is called Liquid Glass — a refined take on Apple's existing translucency system that applies dynamic blur, refraction, and ambient light simulation to interface elements. In practice, it makes the UI feel more responsive to real-world lighting conditions, with window chrome and modal sheets appearing to react to the surrounding environment. The effect is most visible on the new Dynamic Island interactions and on the rebuilt Siri interface, where conversation bubbles shift subtly as the user moves the device.

For foldable-ready devices — Apple confirmed that a foldable iPhone is in development, though it was not shown today — iOS 27 introduces Split View-style multitasking features designed to take advantage of larger unfolded screens, with Siri able to operate across both display halves simultaneously as a persistent side panel.

Privacy Architecture: The Apple Difference

Apple devoted significant keynote time to explaining how the new Siri preserves user privacy despite routing queries through Google's cloud infrastructure. The architecture involves several layers: on-device processing for personal data like contacts, calendar, and messages; a privacy relay that strips identifiers before cloud requests are sent; and contractual agreements with Google that prohibit training on Apple user data. Apple is also offering a full "Private Siri" mode that restricts all processing to on-device models — with reduced capability — for users who decline any cloud processing.

Whether these safeguards satisfy privacy regulators in the EU and elsewhere remains to be seen. The arrangement creates a novel legal question: when a user asks Siri a question that involves their personal data, and the answer is generated by Google's model via Apple's relay infrastructure, who bears liability for the response and how is the data governed? Apple's legal teams have clearly anticipated these questions — the company preemptively published a detailed technical paper on the architecture alongside the keynote — but the regulatory picture will unfold over the coming months as data protection authorities in multiple jurisdictions examine the arrangement.

🔬 TITS Research Perspective

Today's WWDC keynote is a landmark event for TITS researchers studying both AI deployment architecture and the competitive dynamics of the AI industry. Apple's decision to license rather than build illustrates the growing divergence between companies with genuine frontier model capabilities (Google, Anthropic, OpenAI) and those that will integrate those capabilities into products and distribution networks — a pattern TITS analysts have been calling the "model layer vs. application layer" divide. The AI Extensions multi-model framework is also a live experiment in what happens when users gain genuine choice between AI models at the OS level: it creates measurable data on which models users prefer for which task types, with significant implications for how AI labs compete and position their products going forward.

Tim Cook's AI Legacy and What Comes Next

In his closing remarks, Cook was characteristically measured about the significance of the moment. He framed the new Siri not as a capitulation but as an evolution of Apple's longstanding philosophy of choosing the right technology for the right job, drawing a line from Apple's decision to switch CPU architectures multiple times over the company's history to today's decision to use Google's AI. "We don't win by building everything ourselves," Cook said. "We win by building the best experience." It was, by most accounts, a clean exit line for a 14-year CEO tenure that reshaped how the world interacts with personal technology.

John Ternus, who will take over on September 1, was present at the keynote but did not speak. His tenure will begin with an Apple that has finally committed to a clear AI strategy, a stable hardware roadmap anchored by the M-series chips, and a software platform now formally open to the multi-model AI future. The foldable iPhone — expected to be Ternus's first major hardware announcement — will test whether that new software foundation can carry a new device category. Meanwhile, the immediate question for iOS 27 users is simpler: after years of disappointment, does the new Siri actually work? Developers will have their answers within hours of receiving Beta 1. The broader market will find out this fall.

For the AI industry as a whole, today's keynote is a signal that the platform layer has arrived. With Google powering Siri and Claude, Gemini, and ChatGPT all accessible as first-class OS-level assistants on the world's most profitable consumer device, the AI model wars have entered a new phase — one fought not just on benchmarks and capability, but on distribution, integration quality, and the everyday experiences of over a billion users.