Apple’s AI software will live or die by its privacy promise

As expected, yesterday WWDC Keynote It was mostly about artificial intelligence. Also predictably, Apple tried to turn its late arrival into a sales pitch: it didn’t rush into AI because it was taking its time to do things right. In this case, “right” means “more privacy than anyone else.” It’s a good idea – the question is how successful it will be.

the Apple’s new intelligence features The updated Siri AI is designed to work across iPhone, iPad, Mac, Apple Watch, and Vision Pro. There’s a dedicated Siri AI app, with a chatbot-esque ChatGPT experience, new AI-powered camera and photo editing features, and the beginnings of a proxy experience that will allow Siri AI to interact with apps and other software on your iPhone, iPad, or Mac.

Whichever device you access the new AI from, Apple says the processing will be roughly the same: queries will be handled on the device where possible, and in its secure private cloud computing system when that’s not possible. Apple says your data will not be stored, will only be used to fulfill your request, and will not be accessed by Apple or anyone else. Conversation history will be kept in the new Siri AI app, but only on the device and in an end-to-end encrypted iCloud account.

This structure in itself is not new – Private cloud computing has been announced Along with the initial launch of Apple Intelligence in 2024. But two years later, two things have changed. First, it is undeniable that Apple is behind all competitors in the field of artificial intelligence, even after yesterday’s announcements. This makes its privacy offering more important than ever as a way to differentiate itself from the competition. It’s the second change that makes it more difficult: partly because it’s so late, Apple is now working with Google and Nvidia to power its own AI operations.

Instead of designing Apple’s new cloud AI models entirely in-house, they rely on Google Gemini. Meanwhile, Private Cloud Compute has expanded beyond Apple’s data centers, running on Google Cloud systems using Nvidia GPUs, Intel CPUs, and Google Titan chips.

This is a noticeable change. When private cloud computing was first announced, Apple emphasized that it was specifically designed to run on an Apple processor, with a robust supply chain that includes extensive security checks and validation checks on each server before it hits shelves. However, Apple cannot control the supply chains of Google, Intel, and Nvidia. instead of, Apple now maintains A “cryptographically verifiable ledger attached only” to all Google Cloud devices used in Private Cloud Compute and “retains full control” of the software. Apple claims the resulting system has the same “extraordinary security and privacy properties” as before, though skeptics may point out that the longer supply chain introduces vulnerabilities that didn’t exist before.

However, Apple can credibly claim that its approach to AI prioritizes privacy that the AI ​​giants lack. He was not shy about making this point. “Some seem to be racing ahead, seemingly pursuing AI for AI’s sake, without clear regard for the people, all of them, that it is ultimately meant to serve,” Senior Vice President of Software Engineering Craig Federighi said during the keynote. He went on to explain that Apple Intelligence was instead designed “with privacy in mind at every step.”

In fact, Apple’s privacy solutions aren’t entirely unique, though they remain the most comprehensive in the industry. last year Google announced private AI computingwhich we called at the time “nearly identical” to the Private Cloud Compute (right down to the name), which runs on “one seamless Google stack” powered by its own internal TPUs. The difference is in the application: While Apple uses private cloud computing for every AI-driven query that can’t be run on the device, Google has been less specific about when it uses private AI computing — and when it doesn’t. The company says it’s used in the AI-powered Magic Cue and Recorder app on Pixel phones, but didn’t say if it’s used for every Gemini query made from a Google phone, or if there are any equivalent protections for using Gemini on other platforms.

In reality, Google says Gemini collects large amounts of data by default: prompts you send, files you share, and recordings of spoken conversations. Google also collects data about the content Gemini creates for you, the tasks it performs, and information from and about the apps, browsers, and devices you use with it. By default, your chat history with Gemini is stored for 18 months and then deleted, although you can reduce this to as little as 72 hours.

It’s a similar story with other major AI companies. OpenAI It says it collects the prompts and content users upload to ChatGPT, as well as various types of location and device information. Chats are used as training data by default, although you can disable this. Anthropic It brings together pretty much the same things when using Claude, although it does note that it deletes the audio recordings of spoken prompts, keeping the transcripts. Claude also uses your data virtually for model training, and says it is kept in a “de-identified” form for up to five years.

In contrast, the Apple Intelligence Privacy Policy — last updated in 2025, before the most recent announcements — states that Apple collects only “limited information” about private cloud computing requests, such as the size and how long they take to complete, but no information about the content or outcome of the request. Naturally, Apple has an advantage here: it doesn’t need to collect your data to train its models, because Google has already handled that with that it User data. Apple’s partnership with Google may raise concerns about how it maintains user privacy, but it may also be precisely how well it can deliver functional AI without giving up that privacy.

The more limited virtual data collection is what ultimately makes Apple’s privacy promise stand out. Even if expanded private cloud computing turns out to not be as secure as it once was, Apple still claims that it collects far less data than any of the AI ​​alternatives. For many Apple users, it may be worth waiting a few months — or years — for Apple Intelligence to finally arrive.