---
title: "Is Siri AI just Gemini? No, and here is what Google actually built"
description: "Apple's own documents explain who built what, where your Siri requests are processed and why Apple says Google cannot read them."
url: https://mactoreality.com/is-siri-ai-gemini
published: 2026-10-14T09:00:00Z
updated: 2026-10-22T12:21:55.000Z
author: "Mac to Reality Newsdesk"
category: Guides
publisher: Mac to Reality
apple_heat_impact: 0
---

# Is Siri AI just Gemini? No, and here is what Google actually built

*Apple's own documents explain who built what, where your Siri requests are processed and why Apple says Google cannot read them.*

**Key points**

- Siri AI runs on Apple Foundation Models built with Gemini technology and refined using outputs from Gemini frontier models.
- Four of the five models run on your device or Apple's own servers; only AFM 3 Cloud Pro runs in Google Cloud.
- Apple says Google cannot read requests, and said its full set of Google Cloud protections would ramp up gradually over the summer.

No. Siri AI is not Google Gemini with an Apple logo on it. It runs on Apple’s third generation of Apple Foundation Models (AFM 3), a family of five models that Apple says were “custom-built in collaboration with Google” using the technology behind Gemini. Apple handles the post-training, architecture, compression and hardware optimisation. The models run on your iPhone, iPad or Mac, or on Apple’s Private Cloud Compute servers, and Apple says Google cannot read your requests. Gemini outputs helped refine the models, but the finished models, the assistant and the privacy rules belong to Apple.

## Where the “Siri is just Gemini” myth comes from

On 12 January 2026, Apple and Google published a joint statement announcing a “multi-year collaboration” and saying that “the next generation of Apple Foundation Models will be based on Google’s Gemini models and cloud technology”. It added that Apple had “determined that Google’s AI technology provides the most capable foundation for Apple Foundation Models”.

Headlines shortened that to “Gemini will power Siri”, and on social media it became “Siri is just Gemini now”. The first version is a simplification. The second is wrong, and Apple’s own technical documents, published at WWDC in June and expanded when Siri AI launched on 14 September, explain why.

## What “built on Gemini” actually means

Apple’s machine learning research post describes how the models were made. All five “shared a common initial foundation” before each was specialised for its job, and pre-training ran on “the latest generation of cloud TPU accelerators”, which are Google’s own AI chips. That shared foundation is where Gemini comes in. At WWDC on 8 June, Amar Subramanya, Apple’s vice president of AI, described the four models that run on Apple silicon as “custom built for Apple Silicon, trained using proprietary data with reinforcement learning and refined using outputs from Gemini frontier models”. In other words, Gemini acted as a teacher whose outputs helped shape Apple’s models, rather than Gemini answering your questions. Apple’s security team puts it plainly: Apple “collaborated with Google to leverage the technologies behind its Gemini family of models to build the next generation of Apple Foundation Models”.

The rest of the work is Apple’s:

- **Post-training:** Apple combined supervised fine-tuning with multi-stage reinforcement learning to shape how the models answer.
- **Apple’s own architecture:** the largest on-device model, AFM 3 Core Advanced, uses a technique called Instruction-Following Pruning that Apple describes as “developed by Apple researchers”. It keeps the full 20-billion-parameter model in flash storage and loads only the parts a request needs.
- **Compression and hardware:** Apple compressed the models with quantisation-aware training and optimised four of the five to run on Apple silicon.

A useful comparison is a car built on another manufacturer’s platform. The engineering underneath is shared, but the car that comes out has its own body, software, badge and warranty. Siri AI is Apple’s car on a Gemini platform.

## The five models behind Siri AI, and where they run

| Model | Where it runs | What it does |
| --- | --- | --- |
| AFM 3 Core | On your device | A 3-billion-parameter model for everyday on-device tasks |
| AFM 3 Core Advanced | On your device (most capable Apple silicon only) | A 20-billion-parameter multimodal model that activates 1 to 4 billion parameters at a time, used for expressive voices and more accurate dictation |
| AFM 3 Cloud | Private Cloud Compute on Apple silicon | Apple’s “server-side workhorse” for requests too big for the device |
| ADM 3 Cloud (Image) | Private Cloud Compute on Apple silicon | Image creation and editing, including the new Image Playground |
| AFM 3 Cloud Pro | Private Cloud Compute on NVIDIA GPUs in Google Cloud | The most demanding work, such as agentic tool use and complex reasoning |

Only the last model runs in Google’s data centres. Everything else runs on your device or on Apple’s own servers.

## Does Google see what you ask Siri?

According to Apple, no. Requests handled on the device never leave it. Requests that go to the cloud are processed by Private Cloud Compute, which Apple says “ensures that user data is never stored or shared with anyone, including Apple”. That applies to AFM 3 Cloud Pro as well, even though its servers sit in Google Cloud.

Apple’s security research blog lists how that works on Google’s hardware:

- The same five rules as the rest of Private Cloud Compute apply: stateless computation, enforceable guarantees, no privileged runtime access, non-targetability and verifiable transparency.
- Apple keeps “complete control” over the server software, and Apple devices only trust software that Apple has cryptographically approved.
- The servers use NVIDIA Confidential Computing, Intel’s TDX and Google’s Titan chip, and the most sensitive parts are checked against at least two independent roots of trust.
- Apple keeps a cryptographically verifiable, append-only record of every Google Cloud machine in the fleet.
- Apple publishes the server software for inspection and gives security researchers access to live nodes through the Apple Security Bounty programme.

One caveat is worth knowing. When Apple announced the Google Cloud setup in June, it said the system would be “gradually ramping towards the complete set of protections throughout the summer preview period”. The promise is that outside researchers can check this, rather than taking Apple’s word for it.

> **What Google does, and doesn’t do, in Siri AI**
>
> - **Google does:** supply the Gemini model technology and outputs Apple used to build and refine its models, the TPU chips used for pre-training, and the Google Cloud data centres that host AFM 3 Cloud Pro.
> - **Google doesn’t:** run the Siri assistant, decide what Siri can do, or get readable access to your requests, according to Apple.
> - **Training data:** Apple says it does “not use our users’ private personal data or user interactions” to train its foundation models.
> - **Branding:** nothing in Siri AI is labelled Gemini, and asking Siri a question does not open Google’s Gemini app.

## Siri AI is not the Gemini app

Google’s Gemini app for iPhone is a separate product. It is Google’s chatbot, you sign in with a Google account, and Google’s privacy policy applies. Siri AI is built into Apple’s operating systems, and Apple’s descriptions of it don’t involve a Google account or the Gemini app at any point. Two iPhone owners could use both side by side, and their questions would go to two different companies under two different sets of rules.

There is one more place where Google technology shows up. In its Siri AI announcement, Apple said it will support the SynthID standard, which Google created to mark images made or edited with AI. That is a labelling standard for pictures from Image Playground and Photos, not Gemini answering your questions.

## Why Apple built Siri AI this way

The joint statement gives Apple’s reason in one line: after “careful evaluation”, Google’s AI was “the most capable foundation”. The same statement promised that “Apple Intelligence will continue to run on Apple devices and Private Cloud Compute, while maintaining Apple’s industry-leading privacy standards”.

In practice, that lets Apple start from a model family that was already strong, and spend its own effort on what it controls: shrinking models to run on iPhones, wiring Siri into apps and personal data, and the privacy system around the cloud. Apple still does its own model research, as the Core Advanced architecture shows.

## What this means for you

Siri AI’s answers benefit from Gemini-class technology, but the privacy terms are Apple’s, and Apple decides where and when Siri AI is available. Siri AI is not available initially in the EU on iPhone, iPad and Apple Watch, for example, even though Google’s Gemini app is. On the Mac, Apple’s EU restriction does not apply. Our guide to Siri AI supported devices, languages and countries covers which iPhones, iPads and Macs get it, and where.

If you see a post saying Siri is “just Gemini” or that “Google reads your Siri requests”, the accurate answer is shorter: Siri AI is built on Gemini technology, but it is Apple’s model, running on Apple’s terms.

## Frequently asked questions

### Is Siri AI the same as Google Gemini?

No. Siri AI runs on Apple Foundation Models (AFM 3), a family of five models that Apple says were custom-built in collaboration with Google using the technology behind Gemini. Apple says they were refined using outputs from Gemini frontier models, and Apple handles post-training, compression and hardware optimisation itself. The models run on your device or on Apple's Private Cloud Compute.

### Does Google see what I ask Siri AI?

According to Apple, no. On-device requests never leave your device. Cloud requests are handled by Private Cloud Compute, which Apple says never stores or shares user data with anyone, including Apple. That also applies to the one model that runs in Google Cloud, where Apple controls the software and publishes it for security researchers to inspect.

### Which parts of Siri AI run on Google's servers?

Only AFM 3 Cloud Pro, Apple's most capable model for agentic tool use and complex reasoning. It runs on NVIDIA GPUs in Google Cloud as part of Private Cloud Compute. The other four models run on your device or on Apple's own Apple silicon servers.

### Do I need a Google account or the Gemini app to use Siri AI?

No. Siri AI is built into Apple's operating systems, and Apple's descriptions of it don't involve a Google account or the Gemini app. Google's Gemini app for iPhone is a separate product with its own sign-in and Google's own privacy policy.

### Are my Siri requests used to train the models?

Apple says it does not use users' private personal data or user interactions when training its foundation models. The models were trained on public data, licensed data, open-source data, data from dedicated studies and synthetic data.

### Why did Apple work with Google instead of building everything itself?

In the joint statement of 12 January 2026, Apple said that after careful evaluation, Google's AI technology provided the most capable foundation for Apple Foundation Models. Apple still does its own model research, such as the architecture of its largest on-device model, and builds the privacy system and the Siri experience itself.

## Sources

- [Apple and Google: Joint statement from Google and Apple (12 January 2026)](https://blog.google/company-news/inside-google/company-announcements/joint-statement-google-apple/)
- [Apple: Introducing the third generation of Apple's foundation models](https://machinelearning.apple.com/research/introducing-third-generation-of-apple-foundation-models)
- [Apple: Expanding Private Cloud Compute](https://security.apple.com/blog/expanding-pcc/)
- [Apple: Siri AI, a profoundly more capable and personal assistant, is here](https://www.apple.com/newsroom/2026/09/siri-ai-a-profoundly-more-capable-and-personal-assistant-is-here/)