Apple Intelligence at WWDC 2026 Bets on Gemini and Raises a Privacy Question

Apple held its WWDC 2026 keynote on June 8, and Apple Intelligence was the centrepiece -- a long-overdue Siri overhaul that transforms the assistant from a glorified search shortcut into something that can hold a conversation, read the screen, and take multi-step actions across apps. But buried inside the Apple Intelligence WWDC 2026 announcements was a detail worth examining: for demanding queries, Apple is routing requests through Google Gemini's infrastructure.

What Apple Actually Announced at WWDC 2026

The new Siri AI can retain context across prompts, read what is currently on your screen, access your files and past conversations, and execute multi-step instructions across apps -- booking a table while checking your calendar, for example. According to Apple's keynote transcript published June 8, Siri can now "take actions across apps" in a meaningfully broader sense than its previous voice-shortcut model.

Visual Intelligence is also new: point your iPhone camera at an object, a restaurant menu, or a sign, and Siri can describe, translate, or respond to it directly from the Camera app.

iOS 27 and macOS 27 developers will get new APIs for registering app actions that Siri can invoke from natural language -- a meaningful shift in how apps get discovered and used on Apple platforms.

The Two-Tier Model Stack Under Apple Intelligence

Here is the architecture Apple quietly introduced. Apple Intelligence now operates on two tiers:

• On-device Apple Foundation Models (AFM) -- local processing, fast, private, for lightweight tasks.
• Cloud models via Private Cloud Compute -- for complex queries, Apple has partnered with Google to integrate Gemini-based models, according to multiple post-keynote reports including Business Standard's WWDC 2026 coverage.

Apple also introduced AFM Cloud Pro, its own cloud-tier foundation model, described during the keynote as matching Gemini Frontier-class quality for the most demanding use cases. The practical result is that a single Siri request can touch your device's local chip, Apple's own cloud model, or Google's Gemini stack, depending on how complex Apple's routing logic judges the query to be.

The Privacy Trade-off

Apple's Private Cloud Compute is designed to be more privacy-preserving than a generic cloud API. Apple has published architecture documentation showing cryptographic attestation of the processing environment, and queries are not supposed to be retained after a session. That is meaningfully better than many cloud AI services.

But it is still server-side processing. A complex request leaves your device and travels through infrastructure that, at least in part, involves Google. For most users the trade-off is easy: smarter answers are worth it. For users who need stricter control over where queries go -- clinicians, legal professionals, or privacy-conscious individuals -- it is a real consideration, and one Apple has not made easy to audit from inside the Settings app.

What Developers Should Know

iOS 27 and macOS 27 expose new Siri intent APIs that let developers register natural-language app actions. Your app could be called by the user's Siri AI without the user opening it directly, which changes how discoverability works on Apple platforms. It also means Siri can silently hand off tasks to third-party apps without surfacing which app handled the request.

The new Siri AI features will not be available in the EU or China at launch, per Apple's own announcement. EU AI Act enforcement and China's separate AI regulations both add friction to rolling out features tied to external cloud providers such as Gemini -- a dynamic that is unlikely to ease quickly.

What the Gemini Partnership Signals for the AI Market

Apple integrating Gemini into Apple Intelligence tells you something about the AI market's structure in mid-2026. Even a company with Apple's engineering resources and its own active foundation model research is choosing to license frontier-class models from a competitor rather than train everything in-house. A handful of providers -- Google, Anthropic, OpenAI -- appear to have pulled far enough ahead on training scale that even platform companies are routing their most demanding workloads through them.

For everyday users, this means the model answering your question may have nothing to do with the company that made your device. Multi-model routing is already the norm, and tools that give users direct control over which model handles their queries -- and keep those API keys in the user's hands -- become more relevant as the number of undisclosed routing layers grows. That is the premise behind bring-your-own-key tools like ByteChat.

The new Siri AI rollout to U.S. English users is expected later in 2026; Apple has not published a specific date.

Frequently Asked Questions

Is the new Siri AI from WWDC 2026 available now?

Not yet. Apple announced the overhauled Siri AI at WWDC on June 8, 2026, but availability for U.S. English users is expected later in the year as part of iOS 27. The EU and China will not receive the update at launch due to regulatory requirements.

Does Apple Intelligence use Google Gemini?

According to multiple post-keynote reports and Business Standard's WWDC 2026 coverage, Apple has partnered with Google to integrate Gemini-based cloud models for demanding tasks via Apple's Private Cloud Compute. Simpler, on-device tasks use Apple's own Apple Foundation Models. Apple has not fully documented which query types route to which model tier.

What is Apple's AFM Cloud Pro?

AFM Cloud Pro is Apple's own cloud-tier foundation model for the most demanding tasks, described at the WWDC 2026 keynote as matching Gemini Frontier-class quality. It runs via Apple's Private Cloud Compute rather than on-device, and it is separate from the Gemini cloud-model integration that handles some other complex queries.