Current status as of June 30, 2026: Apple's WWDC 2026 keynote reframed Siri as "Siri AI" and confirmed what observers had speculated since 2024: the new Siri hands off work to Google Gemini for the supplemental processing that Apple's in-house models do not handle. The launch was covered in detail by MacRumors and confirmed in Apple's official newsroom post. Siri AI will not be available in the EU or China at launch. The general-availability date has not been announced as of June 30, 2026.
Quick facts
What happened
Apple's WWDC 2026 keynote reframed Siri as "Siri AI" — and then quietly confirmed what most observers had been speculating about since 2024: the new Siri hands off work to Google Gemini for the supplemental processing that Apple's in-house models cannot handle.
That one partnership is the most consequential line in the entire keynote. It changes what the iPhone Pro is, what it is good at, and — for the narrow but important class of user that cares about it — what the device is *not* good at anymore.
All dates and capabilities below are cross-checked against Apple's official Siri AI announcement, MacRumors' WWDC coverage, and the broader Gemini 3.5 launch at Google I/O 2026.
What "Siri AI" actually does
The iOS 27 Siri overhaul is bigger than the marketing suggests. The new Siri is no longer a single-assistant abstraction with a fixed backend. It is a routing layer that decides per-request whether to:
- Handle the request locally with Apple's on-device Foundation Models.
- Hand off to Apple's Private Cloud Compute for higher-complexity work.
- Hand off to Google Gemini for the class of work that Apple's models do not handle well — long-context synthesis, web-grounded reasoning, real-time world knowledge, and multimodal tasks that span text, image, audio, and video.
Per Apple's newsroom post and the MacRumors WWDC recap, the new Siri ships with these capabilities:
- Multi-turn conversation. A real conversational thread that holds context across turns and across apps.
- Real-time world knowledge. Search-grounded answers that draw on current web information, not just the model's training cutoff.
- Personal context across apps. Siri can pull from messages, mail, photos, calendar, and files to answer questions that depend on the user's data — and it does so with on-device context for the parts that need to stay private.
- Camera mode. A new Siri mode inside the iPhone Camera app lets users point the camera at objects and interact with visual content directly.
- Voice engine revamp. A more expressive voice with micro-adjustable settings.
- Dedicated Siri app. A standalone Siri app where users can scroll through past conversations, synced via iCloud for cross-device continuity. Coming to watchOS too.
- visionOS 3D visualization. On Vision Pro, Siri gains a 3D visualization that users can place in their physical space.
- Mac integration. Spotlight gets Siri integration, plus right-click context-menu access throughout macOS.
The piece that Apple did not put in the keynote but was confirmed by Gemini's launch materials at Google I/O 2026 is the Gemini handoff. For workloads that Apple's Foundation Models cannot handle in-house, Siri routes to Gemini 3.5 Flash — which Google launched at I/O as positioned for high-throughput output and the default model for the Gemini app and Google Search's AI Mode.
Why Apple did not build this in-house
The most direct framing of the Apple-Gemini partnership is that Apple's on-device Foundation Models are not at the frontier in 2026. Per the Google I/O 2026 coverage, Gemini 3.5 Flash outperforms Apple's in-house models on the workloads Apple needs Siri to handle at scale — long-context synthesis, real-time web grounding, multimodal reasoning.
Apple had three options:
- Catch up in-house. Build Foundation Models large enough to match Gemini 3.5 Flash on long-context and web-grounded tasks. This is a multi-year, multi-billion-dollar investment that competes directly with Google, OpenAI, and Anthropic for the same scarce compute and research talent.
- Stay smaller and route. Keep Apple's Foundation Models focused on the workloads they are good at (on-device personal context, short-form conversational AI, app-level integration), and route the rest to a frontier partner. This is what Apple actually shipped.
- Acquire a frontier lab. A multi-billion-dollar acquisition of an OpenAI competitor. Apple has not gone down this path as of June 30, 2026.
The choice Apple made — option 2 — is the most pragmatic, but it has consequences. The iPhone Pro is now a device whose AI capability is bounded by two things: Apple's Foundation Models for the local path, and Google's Gemini 3.5 Flash for the cloud path. Apple has retained control of the integration layer (the routing, the UI, the personal-context handling), but it has conceded the frontier-model race to Google for the cloud-side workloads.
What the EU and China exclusion means
Per Apple's launch materials, Siri AI will not be available in the EU or China at launch. The reasons are not the same in both regions:
- EU. The EU AI Act and the Digital Markets Act impose specific requirements on AI systems deployed by "gatekeeper" platforms. Apple is positioning Siri AI as a feature that does not yet meet those requirements, and the rollout is gated on regulatory compliance work that is in progress.
- China. Siri AI requires cloud-side processing through Google's Gemini, and Google does not operate Gemini in China. Apple would need either a Chinese cloud partner for the equivalent capability (Baidu, Alibaba, or DeepSeek are the candidates) or a fully on-device implementation that does not depend on Google.
For an enterprise IT team rolling out iPhones to staff in the EU or China, the practical implication is that Siri AI is gated by region, and the rollout timeline for those regions is open-ended as of June 30, 2026.
Source conflict notes
A few points where public reporting and official documentation do not fully align as of June 30, 2026:
- Gemini model version. Apple newsroom confirms the Gemini partnership but does not specify Gemini 3.5 Flash in Apple's text; Google I/O 2026 materials identify Gemini 3.5 Flash as the partner. Readers should verify the latest Apple developer documentation for the specific Gemini SKU before procurement decisions.
- Routing rules. Apple has not published a routing-rule spec that maps request categories to local-vs-Gemini-vs-Private Cloud Compute. The architecture description is qualitative.
- EU/China GA date. Apple has not committed to a date. The rollout is described as a regulatory and partner-coordination dependency.
The architectural pattern: AI router, not AI model
The deeper story here is not Apple-specific. It is the architectural pattern that the entire industry is converging on in 2026: the AI router. Instead of building one model that does everything, the platform ships a routing layer that picks the right model per request.
This is the same pattern OpenAI ships with the Apps SDK, where the model layer is wrapped by MCP servers that provide tools and resources. It is the same pattern the OpenAI Apps SDK MCP server concepts document. Anthropic's Claude Cowork and Claude Code follow the same model — a routing layer that picks between Claude Opus, Sonnet, Haiku, and external tool providers per task.
For an enterprise IT team, the implication is that the model layer is no longer the strategic surface. The integration layer — the routing, the personal-context handling, the audit logging, the data governance — is the strategic surface. Apple owns the iOS integration layer. Google owns the frontier model layer. The two are now coupled at the API, and the user experience is the result of how Apple decides to route.
What this means for the iPhone Pro buyer
If you are buying an iPhone Pro in late 2026 — and especially if you are an executive buyer with specific privacy, regulatory, or operational requirements — three things change:
1. Siri is no longer a single-vendor product. A Siri query that hits the cloud path is now a Google Gemini query, governed by Google's terms and Apple's privacy contract. If your enterprise has a specific position on Google's data handling — for GDPR reasons, for competitive reasons, for sector-specific reasons — that position now applies to Siri AI requests that get routed to Gemini.
2. The on-device path is the privacy path. Apple's Foundation Models run locally for the workloads they handle. The workloads they *don't* handle — long-context synthesis, real-time web grounding, multimodal reasoning — go to Gemini. For a user who wants to keep a query fully local, the on-device path is the only path, and not every query is eligible for it.
3. The regional rollout is your rollout. EU and China buyers will get Siri AI later. If you are an enterprise with EU or China staff, the iOS 27 rollout is not a single date — it is a per-region date that depends on regulatory work Apple has not finished as of June 30, 2026.
For most users, none of this matters. Siri AI will feel like a smarter Siri. For the narrow class of user that cares about which vendor handles the cloud-side inference, the architecture is now explicit, and the answer is "Apple for the local path, Google for the cloud path."
What changed since the last update
Compared to the pre-WWDC Siri:
- Architecture. Siri is now a routing layer, not a single-model assistant.
- Cloud dependency. Frontier-class queries route to Google Gemini (3.5 Flash).
- Regional availability. EU and China are excluded at launch.
- Capability surface. Multi-turn conversation, personal context, and camera mode are added.
FAQ
Does iOS 27 Siri AI use Google Gemini? Yes, for the supplemental processing workload. Apple Foundation Models handle on-device tasks; Apple's Private Cloud Compute handles higher-complexity in-house work; Google Gemini 3.5 Flash handles long-context synthesis, real-time world knowledge, and multimodal reasoning. Apple has not published the routing-rule spec.
Why is Siri AI not available in the EU or China at launch? EU: the EU AI Act and the Digital Markets Act impose specific requirements on gatekeeper AI features; Apple is positioning Siri AI as not yet meeting those requirements. China: Gemini is not operated by Google in China, so Apple would need a Chinese cloud partner or a fully on-device alternative.
When will iOS 27 Siri AI launch? Apple has not committed to a specific date. As of June 30, 2026, the GA date is gated on Apple's deployment schedule for the regions where Siri AI is launching.
Can I keep Siri AI queries fully on-device? For some queries, yes — those that Apple's Foundation Models handle. For long-context synthesis, real-time world knowledge, and multimodal reasoning, the queries route to Gemini. Apple has not published which query types are routed where.
What happens to my Apple Intelligence data when Siri AI launches? Apple's launch materials describe personal-context handling with on-device processing for the parts that need to stay private. The full data-handling spec is in Apple's Siri AI announcement.
Sources checked
- Apple — Introducing Siri AI: a profoundly more capable and personal assistant — official launch announcement and capability list
- MacRumors — Apple announces Siri AI at WWDC 2026 — keynote recap with feature-level breakdown
- Google Blog — Google AI updates at I/O 2026 — Gemini 3.5 launch and the Apple partnership framing
- TechRadar — Apple Siri WWDC 2026 audio engineer analysis — engineering details of the voice engine revamp
- Mashable — All the Gemini announcements at Google I/O 2026 — partner-side framing of the Apple-Gemini handoff
- OpenAI Apps SDK — Build with the Apps SDK — comparable routing-layer pattern on the OpenAI side
- OpenAI — Apps SDK MCP server concepts — the developer-side documentation for the AI-router pattern
For a Different Kind of Audience
If your iPhone Pro workflow is "send a message, schedule a meeting, ask Siri to draft an email, ask Gemini to summarize a web page," iOS 27 Siri AI is exactly the right tool and you will appreciate the upgrade. If your iPhone Pro workflow is "an executive conversation that cannot be routed to a third-party cloud for processing — any third-party cloud — no matter what the privacy contract says," a different class of device exists. See the luxury phones with on-device AI assistants guide for hardware that runs frontier-class models without any cloud handoff.
