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

As expected, yesterday Image from WWDC most of it was about AI. Also as expected, Apple tried to turn its late arrival into its selling point: it did not rush to AI because it took time to develop. In this case, “right” means “having more privacy than anyone else.” It’s a good word – the question will be how well it works.

The new Apple Intelligence and the updated Siri AI is designed to work on iPhone, iPad, Mac, Apple Watch, and Vision Pro. There’s a dedicated Siri AI app, with a ChatGPT-esque chatbot, a new AI-powered camera and photo-editing feature, and tech basics that will allow Siri AI to interact with other apps and programs on your iPhone, iPad or Mac.

Whichever device you get from the new AI, Apple says the process will be the same: questions will be answered on the device if possible, and on its own Private Cloud Compute when it’s not. Apple says your data will not be stored, it will be used only for the purposes you want, and it will not be accessed by Apple or anyone else. Conversation notes in the new Siri AI app will be saved, but only on devices and in your iCloud account stored end-to-end.

This architecture is not new – Private Cloud Compute announced along with the first launch of Apple Intelligence in 2024. But two years, two things have changed. First, Apple is undoubtedly behind any competition on AI, even if it announced yesterday. This makes privacy more important than ever as a way to differentiate yourself from the competition. It’s the second change that makes this even more difficult: in part because it’s behind the scenes, Apple is now working with Google and Nvidia to drive its AI work.

Instead of being developed in-house, Apple’s new AI cloud models come from Google Gemini. Currently Private Cloud Compute has expanded beyond Apple’s data centers, running on Google Cloud machines using Nvidia GPUs, Intel CPUs, and Google Titan chips.

That’s a big change. When Private Cloud Compute was first announced, Apple emphasized that it was built to run on Apple’s silicon, with robust integration tools including extensive security scans and validation checks on each server before it joins. Apple cannot control the chains of Google, Intel, and Nvidia. On the contrary, Apple now maintains it a “privately authenticated, self-extending number” of all Google Cloud devices used for Private Cloud Compute and “reserves the full power” of the software. Apple says the upcoming system has the same “amazing security and privacy protections” as before, although skeptics will say that the long delivery process introduces new challenges.

However, Apple can confidently say that its approach to AI prioritizes privacy which is missing from the AI ​​giants. And there was no shame in saying that. “Some seem to be rushing ahead, they seem to be pursuing AI for AI’s sake, not considering the people, all of us, that we can serve,” SVP of software engineering Craig Federighi said at the event. He went on to explain that Apple Intelligence was developed “with privacy in mind at all times.”

Indeed, Apple’s privacy solutions aren’t really that different, though they’re still the most common in the industry. Last year Google has announced Private AI Computewhich at the time we called “equivalent” to Private Cloud Compute (right down to the name), running on “a single Google stack” powered by its internal TPUs. The difference is in the usage: while Apple uses its cloud computing for every AI query that can’t be run on hardware, Google hasn’t mentioned when it uses Private AI Compute – and when it doesn’t. The company says it’s used for Magic Cue and the AI-powered Recorder app on Pixel phones, but hasn’t said whether it’s used for every Gemini query made from a single Google phone, or if there’s a similar security feature for using Gemini on other platforms.

Then, Google says so that Gemini collects a lot of information by default: prompts you send, files you share, and recordings of conversations. Google also collects the information that Gemini makes about you, the services it performs, and information from the apps, browsers, and devices you use. By default, Gemini chat history is kept for 18 months and then deleted, although you can reduce this to 72 hours.

It’s the same story with other big AI companies. OpenAI it says it collects content that users download from ChatGPT, as well as a variety of sites and devices. Chats are used as training data by default, although you can disable this. Anthropic it gathers the same information as when you use Claude, although they note that it removes the spoken audio, and preserves the text. Claude also refrains from using your data to train people, and says it is stored in “de-identified” form for five years.

In contrast, a Apple Intelligence privacy policy – last updated in 2025, before the most recent announcements – it is said that Apple only collects “limited information” on requests for Private Cloud Compute, such as the size and time it takes to complete, but there is no information about the content of the request or the results. Apple says it does not use private data or user communications to train its basic models. Of course, Apple has an advantage here: it doesn’t have to collect your data to train its models, because Google has already done the work. his user data. Apple’s partnership with Google may raise concerns about how it protects users’ privacy, but it may also be about how it provides functional AI without giving up that privacy.

It’s this limited data collection that makes Apple’s privacy promise stand out. Although the expansion of Private Cloud Compute is not as secure as before, Apple still claims that it collects less data than other AI methods. For many Apple users, this may be worth waiting months – or years – for Apple Intelligence to arrive.