Apple split Writing Tools between a device and an audited cloud

A system-wide writing layer, announced 10 June 2024

Apple's announcement of Apple Intelligence introduced Writing Tools as three functions available 'nearly everywhere' a person writes, 'including Mail, Notes, Pages, and third-party apps': Rewrite, which lets someone 'choose from different versions of what they have written, adjusting the tone to suit the audience and task at hand'; Proofread, which 'checks grammar, word choice, and sentence structure while also suggesting edits'; and Summarize, which turns selected text into 'a digestible paragraph, bulleted key points, a table, or a list'. The significant design choice is not any one of these functions, which resemble features already available in standalone apps, but that they are offered as a layer under the operating system's text fields generally, rather than as a feature of one specific application.

Where the processing happens

Apple's announcement states that 'many of the models that power it run entirely on device'. For requests judged too complex for that, the same announcement says Apple built Private Cloud Compute, extending processing to 'larger, server-based models that run on dedicated Apple silicon servers'. Apple's separate technical description of Private Cloud Compute is more specific about what this is meant to guarantee: the system 'uses that data only to perform the operations requested by the user', data 'stays on the PCC nodes that are processing the request only until the response is returned', and, in Apple's words, 'user data is never available to Apple — even to staff with administrative access to the production service'. Apple states it backs this with a published, 'append-only and cryptographically tamper-proof transparency log' of the software running on PCC, and makes the corresponding software images available for outside inspection.

What the announcement leaves for later verification

Apple's own newsroom document says the features 'will be available in beta' with iOS 18, iPadOS 18 and macOS Sequoia 'this fall in U.S. English', without a specific date, and does not state how a person would know, in the moment, whether a request was handled entirely on-device or sent to Private Cloud Compute. The transparency log and public software images address a different audience, security researchers able to verify the published measurements, not an ordinary user deciding whether to summarise a sensitive document. That the guarantee is checkable in principle is a genuine, unusual design choice; whether it is actually checked, by whom, and how often, is not something either document answers.

• Does the text being rewritten or summarised need to stay verifiably on-device, and is that distinguishable from a Private Cloud Compute request in the interface?
• Is the feature available yet in the language and region needed, given the initial U.S. English beta scope?
• Has anyone independently reviewed the published Private Cloud Compute transparency log, or is the guarantee being taken on trust?

Writing Tools is a bet that a system-wide assistant needs a correspondingly systemic answer to where text goes, not just a privacy notice. The architecture Apple describes is more verifiable than a typical cloud AI feature's terms of service, but it still asks a user to trust that verification is happening somewhere, by someone, on their behalf.