Apple Intelligence arrives at WWDC

On 10 June, opening the annual Worldwide Developers Conference from the company’s park in Cupertino, Apple spent the better part of an hour on a subject it had spent years sidestepping: artificial intelligence. The product has a name. Apple Intelligence, the company said, is a personal intelligence system that lives across the iPhone, the iPad and the Mac, understanding what is on screen, writing on the user’s behalf, and organizing the mess of messages, photographs and documents that a phone accumulates over a year.

The announcement had long been expected and still managed to surprise, because the usual way Apple does these things is to ship quietly and let its competitors do the loud announcing. Here the framing was deliberately different. Apple Intelligence is not a chatbot you visit in a web page; it is a layer underneath the operating system, available wherever you happen to be, so that writing an email, rewriting a too-formal notification, or finding the exact photograph of a dog at the beach all happen without ever switching apps. The features arrive with iOS 18 on the phone, iPadOS 18 on the tablet, and macOS Sequoia on the Mac, and they are pitched as working with the system you already use rather than asking you to learn a new one.

The most striking partnership news was reserved for the assistant. Apple confirmed an integration with OpenAI so that ChatGPT, powered by GPT-4o, is available on tap inside the system itself; when a request needs more than the on-device model can manage, the phone can hand it over, and a user does not even need an OpenAI account to use the free tier. The company was careful to say the arrangement is not exclusive and that other models will be added in time, which reads less as enthusiasm and more as a hedge against being locked into a single supplier.

The harder argument was the privacy one, because that is the terrain Apple has always chosen to fight on. The design splits the work in two. Simpler requests are handled on the device itself, where the data never leaves the phone. Anything that needs a larger model is sent to something Apple calls Private Cloud Compute, a cloud system running on Apple’s own silicon, built so that the company says it cannot read what is sent to it, and so that the computation can attest to its own integrity. The pitch is that large-language models do not have to mean sending your life to an unknown server; Apple wants to have it both ways, with the convenience of a big model and the seamlessness of a personal one.

The real test is timing. Apple Intelligence is a promise for later this year, and it will run only on the newest hardware, with a list of supported phones that begins well above the budget line. For a company that made its name on software that runs everywhere, arriving late to a field everyone else entered in 2023 is its own kind of story. On 10 June, at least, the keynote made the company sound ready to stop watching the war from the balcony.