Why this matters
Apple Intelligence is no longer a feature you go and find. It now reads your notifications, drafts your replies, sorts your photos and answers through Siri, so the question of what it does with your data is the question of what your phone does with your day.
Apple built its name on the idea that what happens on your iPhone stays on your iPhone. Artificial intelligence strains that idea, because the most capable models are too large to run on a phone and have to live on servers somewhere. Every AI company faces the same problem. What differs is the answer, and Apple’s answer is more considered than most, with one new wrinkle that landed this month.
At its developer conference in June, Apple announced that the next version of Siri will be powered by a model built together with Google, using the technology behind Google’s Gemini. That single fact has done the rounds as “Apple hands Siri to Google”, which is not quite what is happening. To see why, it helps to know that anything you ask of Apple Intelligence is dealt with in one of three places, and only the third involves another company at all.
Most of what Apple Intelligence does happens on your device and never leaves it. Harder requests go to Apple’s own servers, which are built to keep nothing and can be checked by outside researchers. Only a few things reach a third party such as ChatGPT, and only when you allow it. The Google deal sits inside the second place, not a new fourth one.
The Google Gemini question
Start with the news, because it is the reason most people will read this. The worry is reasonable, and the answer is more reassuring than the headline, with real caveats.
Apple and Google have signed a multi-year deal under which Apple’s next generation of in-house models is built on the technology behind Google’s Gemini, and one of those models will power a more capable, more personal Siri. The part that matters for your data is where that model runs. Apple says it runs on Apple’s own Private Cloud Compute, the same servers described later in this guide, and not on Google’s cloud. In Apple’s words, your requests are not logged or shared with Google, and you interact only with Apple’s models, on your device and in Apple’s cloud.
So Gemini is the recipe, not the kitchen. Google’s technology shaped how the model was trained, but the model that answers you is run by Apple, under the same no-retention promise as the rest of Apple Intelligence. Apple’s chief executive has stated plainly that the partnership does not change Apple’s privacy rules.
There are two honest caveats. The first is timing. The rebuilt Siri is not here yet; Apple has said it arrives later in 2026, so today this is a commitment rather than something you can switch on and test. The second is that, to find the room for these heavier models, Apple has begun running Private Cloud Compute on hardware in Google’s data centres as well as its own, a point we return to in the section on the cloud. Apple says its guarantees hold there too, through the same design. Whether that reassures you depends on how much of the design you are willing to trust, which is a fair thing to weigh rather than wave away.
The test will come when the new Siri ships. Independent security researchers will examine the Private Cloud Compute software, as they have before, and what they find is worth more than any launch announcement. If you are cautious, there is no harm in waiting to turn on the new Siri features until that scrutiny has happened.
Where your request goes
Three places, in order of how much they ask of you. The further down the list a request travels, the more deliberately Apple makes you choose it.
On your device
Most of what Apple Intelligence does never leaves your phone. The summaries at the top of a long email thread or a noisy group chat, the writing tools that proofread or reword a message, the sorting that finds the right photo, and many everyday Siri answers are produced by models that run on the device itself. There is no server, no account and nothing to send, which is also why these features work in flight mode.
Apple’s private cloud
Some requests are too heavy for a phone, so they go to Apple’s servers, called Private Cloud Compute. Apple sends only the data relevant to that one request, the servers work on it, the answer comes back, and Apple says nothing is kept. This is where the larger language models live, including the Gemini-trained one coming to Siri. It is Apple’s middle ground: more power than the phone, without handing your life to a company that profits from data.
A third party
A few requests can be passed to an outside service, today meaning ChatGPT, for the kind of open-ended question Siri would rather not attempt itself. This only happens if you have switched the option on, and Siri asks before each hand-off. It is the one place where a company other than Apple sees your request, and it is firmly under your control.
| On your device | Apple’s private cloud | A third party | |
|---|---|---|---|
| who does the work | your iPhone | Apple’s servers | ChatGPT, for now |
| what is sent | nothing | only the request’s data | only what you approve |
| is anything kept | no | Apple says no | per that service’s rules |
| do you choose it | automatic | automatic when needed | you opt in, and are asked |
| works offline | yes | no | no |
What the private cloud promises
Private Cloud Compute is the heart of Apple’s claim, so it is worth knowing what it offers, and where the trust sits.
When a request goes to Apple’s servers, Apple makes a short list of promises. The data sent there is not stored, and is not made readable to Apple. It is used only to answer that one request, after which the result returns to your device and is not retained. The only thing Apple records is a little technical housekeeping, the rough size of the request, which feature asked, and how long it took, which Apple says carries none of the content and is not linked to your Apple Account.
The part that sets this apart from the rest of the industry is verification. Apple publishes the software that runs on these servers and invites security researchers to inspect it, keeps a tamper-evident record of every machine in the fleet, and arranges things so your own device will only trust server software that Apple has cryptographically signed. You cannot run that check at your kitchen table, but specialists can and do, which is a stronger position than taking a company’s word.
The wrinkle is the new one. To run the heavier models behind the rebuilt Siri, Apple has extended Private Cloud Compute onto hardware inside Google’s data centres for the first time, alongside its own. Apple says the same protections apply there, built on confidential-computing chips and a verifiable record of the hardware, and that it keeps full control of the software regardless of whose building the machine sits in. It is a reasonable design. It is also more moving parts and more companies than the original promise, and you are entitled to factor that in.
Apple’s design is unusually open to inspection, but inspection is still done by researchers, not by you. The promise that nothing is kept is enforced by Apple’s own systems and attestations. That is far better than a plain “trust us”, and short of proof you can run yourself. Hold both thoughts at once.
ChatGPT, and being asked first
The one place your request can reach a company other than Apple is the ChatGPT extension, and it is built to ask permission rather than assume it.
Apple lets Siri pass certain questions to ChatGPT, the kind of broad, generate-me-something request that its own models are not aimed at. This is switched off until you turn it on. Once on, Siri still asks before each hand-off, so nothing reaches ChatGPT without a tap from you. When you do allow it, Apple hides your IP address so the request cannot be tied to you by network address, and OpenAI does not store the request, unless you have chosen to sign in with your own OpenAI account, in which case its normal terms apply.
The practical upshot is simple. If you never want your questions leaving Apple’s world, leave the ChatGPT extension off, which is how a new iPhone arrives. If you find it useful, turn it on and rely on the prompt before each hand-off to keep you in charge. Either way, the related guide on keeping your AI chats private covers what these assistants remember once you are signed in to them directly.
Take control
Whatever you make of the design, the settings are yours. You can switch the whole thing off, or keep the parts you like and drop the rest. Here is where each control lives.
Open the Apple Intelligence Report
Go to Settings, then Privacy & Security, and open the Apple Intelligence Report. Turn it on and it logs the cloud requests your device makes, so you can see for yourself how often anything leaves the phone and which features send it. It is the honest place to start before you decide what to change.
The single switch
In Settings, open Apple Intelligence & Siri. The switch at the top turns Apple Intelligence off entirely, which stops the summaries, writing tools, image features and the cleverer Siri while leaving the rest of your phone untouched. Leave it on if you want to keep some features and turn off others below.
Check the extension
In the same Apple Intelligence & Siri screen, find ChatGPT under the extensions and confirm it is off if you would rather no request ever reached it. If you keep it on, leave the setting that makes Siri ask before each hand-off in place.
Notification summaries
If the feature that rewrites your incoming notifications into a one-line summary is not for you, open Settings, then Notifications, then Summarize Notifications, and switch it off, for everything or app by app.
Improve Siri & Dictation
A separate, older setting governs whether Apple keeps samples of your Siri and dictation audio to improve its systems. Open Settings, then Privacy & Security, then Analytics & Improvements, and turn off Improve Siri & Dictation if you would rather Apple kept none of it.
Block it in Screen Time
To remove the features for a child’s device, or your own, open Settings, then Screen Time, then Content & Privacy Restrictions, where an Intelligence & Siri section lets you switch off writing tools, image creation and more, and hold them off behind the Screen Time passcode.
You do not have to pick a side. A reasonable middle is to keep the on-device features that never leave the phone, turn off the ChatGPT extension until you want it, switch off Improve Siri & Dictation, and check the Apple Intelligence Report now and then. You get the useful parts and send almost nothing.
So is it private?
Honestly, more than its rivals, and with more caveats than the adverts imply.
The good is real. Most of what Apple Intelligence does happens on your device and never travels. The cloud it does use is built to keep nothing, and Apple opens it to outside inspection in a way no other large AI company matches. If you are choosing between mainstream assistants on privacy alone, this is the strongest design on offer, and the local guide on running an assistant on your own laptop is the only thing that goes further.
The caution is also real. The most powerful new features, the rebuilt Siri among them, are still arriving, so some of the promise is yet to be tested in the wild. The private cloud now runs partly on Google’s hardware, which adds companies and parts to a story that was once all Apple. And every guarantee, however well engineered, ultimately rests on Apple’s own systems doing what Apple says. None of that makes it untrustworthy. It makes it a considered bet rather than a sure thing, and a bet you can shrink at any time by turning features off. Use the parts that earn their place, leave the rest, and check the report.
Common questions
The questions people ask once they realise the phone in their pocket is now an AI device.
Is Siri sending my data to Google now?
Not in the way the headline suggests. Apple and Google signed a deal for the next Siri to use a model built on Google's Gemini, but Apple runs that model on its own Private Cloud Compute servers, not on Google's. Apple says your requests are not logged or shared with Google, and that you interact only with Apple's own models, on the device and in Apple's cloud. The Gemini technology shapes how the model was trained; it is not a pipe to Google. The catch is that this Siri does not arrive until later in 2026, so for now it is a promise rather than something you can test.
What does Apple Intelligence do on the phone itself?
A great deal. Summaries of mail, messages and notifications, the writing tools that rewrite or proofread text, and many Siri requests are handled by models that run on the device, so the content never leaves it. Apple's rule is to answer on the device when it can, and reach for the cloud only when a request is too heavy for the phone.
What is Private Cloud Compute?
It is Apple's name for the servers it uses when a request is too big for your phone. They run on Apple chips, they are built to hold nothing after the answer is sent back, and Apple says the data is never stored or made readable to Apple. Unusually, Apple publishes the server software so that independent researchers can check the promise, rather than asking you to take its word.
Does Apple keep a record of what I ask?
Apple says it keeps no content. When a request goes to the cloud, Apple collects only limited measurements, such as the rough size of the request, which feature was used, and how long it took. It states this does not include the content of the request or the answer, and is not tied to your Apple Account. You can see the cloud requests your device has made under the Apple Intelligence Report in Settings.
Is ChatGPT switched on by default?
No. The ChatGPT extension is off until you turn it on, and even then Siri asks before it sends anything to ChatGPT. When you allow it, Apple hides your IP address and OpenAI does not store the request unless you have signed in with your own OpenAI account. If you would rather it were never offered, leave the extension off, which is how it ships.
Can I check any of these claims myself?
More than with most AI. Apple publishes the Private Cloud Compute software and invites security researchers to inspect it, and it keeps a verifiable record of the servers in the fleet. Your own device only trusts cloud software that Apple has cryptographically signed. You cannot audit it at home, but you do not have to rely only on Apple's marketing either, which is rare in this field.
Which devices have Apple Intelligence?
An iPhone 15 Pro or any iPhone 16, and iPads and Macs with an M-series chip. Older devices do not run it at all. The rebuilt Siri also has a regional catch: Apple has said it will not launch in the European Union on iPhone and iPad at first.
If I turn it off, do I lose anything important?
You lose the AI features, not the phone. Summaries, writing tools, image creation and the cleverer Siri answers stop, while calls, messages, the camera and everything else carry on as before. You can also turn off single features and keep the rest, so you do not have to choose all or nothing.
So should I trust it?
It is the most carefully built privacy design of any mainstream AI, and most of what you do stays on the device. The cloud parts are engineered to keep nothing and can be inspected by researchers, which is more than rivals offer. What remains is genuine trust: the strongest new features are still rolling out, and the cloud now runs partly on Google's hardware under Apple's control. Reasonable people use it with that in mind, and switch off the features they do not want.