Histoire
juillet 14, 2026
Apple Unveils Overhauled 'Siri AI' Powered by Google's Gemini
At its WWDC event, Apple announced a significant overhaul of its virtual assistant, now called 'Siri AI.' The new system, built on Google's Gemini model, will be more conversational, understand personal context, and integrate deeply across Apple's operating systems, though its rollout will be delayed in the EU due to regulatory issues.
Apple is reintroducing Siri as a core pillar of its software strategy, betting that a deeply integrated, privacy-focused assistant can help it catch up in the AI race even as key features are delayed in major markets.
Early groundwork and delays
Apple first outlined a bold “Apple Intelligence” vision and smarter Siri two years ago but struggled to ship it, prompting criticism that the company was falling behind rivals and contributing to a rare yearlong delay for Siri’s AI overhaul.12 That gap culminated in a February 2026 settlement over earlier, undelivered Siri AI promises and fueled speculation that Apple lacked a coherent AI strategy.3
WWDC 2026: Siri AI finally arrives
At WWDC 2026, Apple formally unveiled “Siri AI,” describing it as an “entirely new version of Siri” that is more conversational and capable, with an expressive voice that users can tune by pace and expressivity.4 The assistant is now central to iOS 27, which Apple says brings the “next generation of Apple Intelligence” and an “all‑new Siri,” including a redesigned interface that surfaces responses from the Dynamic Island and a camera “Siri Mode” that can recognize objects and explain them.51
Siri AI ships as a system-wide assistant and as a standalone chatbot-style app that stores and syncs conversation history via iCloud across iPhone, iPad, Mac, Apple Watch, and Vision Pro.46 Users can type, speak, or upload documents and images, and the app presents past sessions much like ChatGPT or Claude.6
Under the hood: Google Gemini and privacy tiers
Behind the scenes, Apple quietly licensed a custom Google Gemini model—reportedly around 1.2 trillion parameters—to rebuild Siri AI, with a deal estimated at about $1 billion per year.3 Apple describes a three‑tier architecture: simple requests run on device, more complex ones on its Private Cloud Compute, and the heaviest reasoning on Google Cloud, with the contract barring Google from training on Apple user data.34 “We believe privacy in AI is non-negotiable,” software chief Craig Federighi said, arguing that “truly helpful AI must be centered around you and your needs.”37
New capabilities across the ecosystem
With this architecture, Siri AI can ground answers in current web knowledge while drawing on “personal context” across messages, emails, photos, and what’s on screen to execute multi-step tasks.23 Apple is extending the same intelligence to apps: Safari now auto-groups tabs and monitors pages, Passwords can update compromised logins in one tap, Messages suggests AI replies, and Calendar creates events from natural language.185 Shortcuts can be built simply by describing the desired workflow, and a revamped Image Playground offers more natural editing and photorealistic generation.87
Market positioning and regional limits
Analysts see the launch as Apple’s largest AI push to date, aimed at quelling fears it was “losing” the AI race and reframing the contest around usefulness and trust rather than raw model size.97 Federighi contrasted Apple with competitors that require users to manage privacy through temporary chats or manual deletion, presenting Apple as “the AI company that’s actually on your side.”79
But the rollout is uneven: Siri AI and some Apple Intelligence features will debut later this year in English only, and will not initially be available in the European Union or China while Apple negotiates with regulators, underscoring how privacy and competition rules now shape global AI strategies.527