September 24, 20268 min read
Does the New Siri Send Your Data to Google? How iOS 27's Gemini-Based Siri Handles Privacy
iOS 27's Siri is built on Google's Gemini and partly runs on Google Cloud. What Apple says reaches Google, what doesn't, and the settings that control Siri AI.
The short answer
Partly, but not in the way most people fear. The new Siri runs on Apple models built with Google's Gemini, and some requests are processed on Google Cloud hardware (Apple Security Research). But that hardware runs Apple's Private Cloud Compute software, not Google's Gemini service, and Apple says personal data there isn't stored or accessible to anyone, including Apple (Apple Newsroom).
So the honest summary is: Google built part of the model and supplies some of the hardware, but Apple says it controls the software your requests run on. What neither company has published is a plain statement about whether Google staff could ever see a request. Here is what is and isn't on the record.
What changed with Siri in iOS 27?
iOS 27 was released on 14 September 2026, and with it came "Siri AI", a rebuilt Siri with its own app that keeps your conversations (9to5Mac; Apple). Apple labels it a beta. It launched in English first, is not available in the EU or China at launch, and more languages arrive with iOS 27.1 in October (Apple Newsroom; Macworld).
Siri AI needs an iPhone 15 Pro or Pro Max, or any iPhone 16 or later, as well as M1-or-later iPads and Macs (Apple Newsroom).
Where does Google come in?
In a joint statement on 12 January 2026, Apple and Google said the next generation of Apple Foundation Models would be "based on Google's Gemini models and cloud technology," and that Apple Intelligence would continue to run on Apple devices and Private Cloud Compute (Google and Apple joint statement).
At WWDC in June, Apple filled in the details. Its models were "custom-built in collaboration with Google and its Gemini models" (Apple Newsroom), and there are four of them (Apple Machine Learning Research):
| Model | Where it runs |
|---|---|
| AFM 3 Core | On your device |
| AFM 3 Core Advanced | On your device (newer iPhones) |
| AFM 3 Cloud | Apple's Private Cloud Compute |
| AFM 3 Cloud Pro | Private Cloud Compute on NVIDIA GPUs in Google Cloud |
The last row is the one that matters for this question. Apple's security team confirmed that Private Cloud Compute (PCC) is expanding "beyond Apple's data centers" to Google Cloud, for "the most demanding tasks, including agentic tool-use and complex reasoning" (Apple Security Research).
You may have seen a figure of $1 billion a year for the deal. That comes from a November 2025 Bloomberg report and has never been confirmed by Apple or Google (TechCrunch).
Can Google see what you ask Siri?
Here is what Apple has actually committed to, in its own words:
- Personal data isn't stored or shared. In PCC, personal data "is not stored nor made accessible to Apple or anyone else" (Apple Newsroom).
- Apple controls the software. "Apple retains complete control over PCC software," and your device will only send requests to servers running software Apple has cryptographically approved (Apple Security Research).
- The hardware is locked down and published. The Google Cloud servers use NVIDIA Confidential Computing, Intel TDX and Google's Titan chip. Apple publishes an append-only ledger of the Google Cloud hardware in use, and security researchers can test live PCC nodes through Apple's bug bounty (same source).
- Computation is stateless. Requests are processed and not kept, which is a core PCC requirement (same source).
What Apple has not published is a sentence saying "Google employees cannot access Siri requests." Its approach relies on attestation and confidential computing rather than trusting the cloud operator. That's a strong design, and Apple invites outside researchers to test it. But it is a technical control, not a written guarantee about Google, and it's worth knowing the difference.
One more distinction: Google executives have called Google Apple's "preferred cloud provider" (9to5Mac), while Apple says Siri runs on device and in PCC. Both are true: PCC now partly runs in Google's data centers.
Is your Siri data used to train Gemini?
Apple says it does not use "our users' private personal data or user interactions" to train its own foundation models (Apple Machine Learning Research).
Whether Google trains Gemini on Siri traffic is not publicly addressed by either company. Some sites claim Google is "contractually prohibited" from doing so, but we found no Apple or Google source for that, so we're not repeating it as fact. Given PCC's stateless design, Google shouldn't have retained data to train on, but that's an inference, not a statement from either company.
Which Siri requests stay on your iPhone?
Requests that the on-device models can handle stay on your iPhone. Heavier requests go to PCC, and Apple says the most demanding ones, "including agentic tool-use and complex reasoning," can go to PCC nodes on Google Cloud (Apple Security Research).
The catch, raised by the EFF, is that there is no visual indicator telling you which path a given request took (EFF). If that matters to you, you have to manage what Siri can access rather than which server it uses. The general trade-off between local and cloud AI is explained in our guide to local-first AI assistant privacy, and in local vs cloud AI models.
How to control what the new Siri can access
These settings come from the EFF's September 2026 guide and other published walkthroughs. Some guides were written during the beta, so check the exact wording on your iPhone, as paths can shift in updates.
| What you want | Where to change it | Source |
|---|---|---|
| Stop the Siri app keeping chats forever | Settings > Siri > Keep Conversations: 30 Days or 1 Year (default is Forever) | Tom's Guide |
| Limit which apps Siri can read from | Settings > Siri > App Access | EFF |
| Stop sharing audio and usage for improvements | Privacy & Security > Analytics & Improvements > Improve Siri & Dictation | EFF |
| Go back to the old Siri | Screen Time > Content & Privacy Restrictions > Siri > Siri Classic, then restart | iDownloadBlog |
| Turn Siri off entirely | Settings > Siri > Turn Off Siri | Cult of Mac |
There is no single switch to turn off all of Apple Intelligence in iOS 27, so you manage it feature by feature (Cult of Mac). The EFF also notes you can't block Siri's on-screen awareness.
Is there a limit on how much you can use Siri AI?
Yes. Apple says Siri AI and its cloud models have usage limits that vary with complexity and demand. When you hit one, a feature returns "after a time out period," and Apple says increased access "will be available for a fee" (Apple Support). Apple hasn't announced a price.
Server-side AI has a running cost, and usage limits are how Apple manages it, as other assistants in the personal AI agent race do with paid tiers. Our guide to running AI locally and offline covers what is possible on your own hardware today.
Common questions about the new Siri and Google
Does the new Siri use Google Gemini?
Yes, indirectly. Apple's new foundation models were built in collaboration with Google and are based on Gemini, according to Apple and Google's January 2026 joint statement. Siri runs these Apple models, not Google's consumer Gemini app or service.
Does Siri send my requests to Google's servers?
Some requests can be processed on servers in Google's data centers, because Apple's Private Cloud Compute now partly runs on Google Cloud. Apple says those servers run only Apple-approved software and don't store your data. Simpler requests stay on your iPhone.
Is my Siri data linked to my Google account?
Nothing Apple has published suggests any link to a Google account. Requests go to Apple's Private Cloud Compute, not to Google's Gemini service, and Apple says personal data there is not stored or made accessible to anyone.
Can I turn off the new Siri AI?
You can switch back to Siri Classic under Screen Time > Content & Privacy Restrictions > Siri, then restart your iPhone, or turn Siri off in Settings > Siri. There's no single switch for all of Apple Intelligence.
Sources
- Joint statement from Google and Apple, 12 Jan 2026
- Apple Intelligence brings powerful AI capabilities into everyday experiences, Apple Newsroom, 8 Jun 2026
- Expanding Private Cloud Compute, Apple Security Research, 8 Jun 2026
- Introducing the third generation of Apple Foundation Models, Apple Machine Learning Research, 8 Jun 2026
- Apple Intelligence usage limits, Apple Support, Sep 2026
- iOS 27, Apple
- iOS 27 now available, 9to5Mac, 14 Sep 2026
- Google execs suggest Gemini-powered Siri will run on Google's servers, 9to5Mac, 4 Feb 2026
- How to limit what Apple's new Siri AI can access, EFF, 18 Sep 2026
- The new Siri app is keeping receipts, Tom's Guide, Jul 2026
- How to use Siri Classic, iDownloadBlog, Aug 2026
- Turning off Apple Intelligence in iOS 27, Cult of Mac, Sep 2026
- iOS 27 features and release, Macworld, Sep 2026
- Apple nears deal to pay Google $1B annually (report), TechCrunch, 5 Nov 2025
The new Siri shows how much of "personal AI" now depends on someone else's cloud. Vyra by Vyraagi takes the opposite default: memory stays on your device, with a local model fallback when you're offline. If that appeals to you, join the waitlist.
Vyra is in closed alpha now, with a Founders Beta ahead of public launch.
Related reading
AI Assistant Privacy: What Local-First Actually Changes
"Local-first" gets used as a privacy claim far more often than it earns one. Here's what running an assistant on your own machine genuinely protects, what it doesn't, and the questions worth asking.
How much memory do you need to run local LLMs in 2026? (And why Mac minis keep selling out)
Weights, KV cache and OS overhead, worked out from bits per weight. A size table for 3B to 70B models and why bandwidth sets tokens per second.
ChatGPT Sponsored Agents, explained: how ads inside AI agents work and what they mean for users
What OpenAI's Sponsored Agents are, who sees ChatGPT ads, what data targeting uses, how to turn ads off, and the open questions about ads inside agents.