Siri AI beta: what it can search, where requests go and which controls matter

Four days changed the question around Apple’s new assistant. On 10 September 2026, a WIRED guide still described Siri AI and iOS 27 as forthcoming. On 14 September, Apple released iOS 27 and began rolling out Siri AI as a beta.

That shift from demonstration to availability makes a list of features less useful than an account of what sits behind them. Apple says Siri AI can draw on messages, mail, photographs, calendars and material displayed on the screen. It can also retrieve information from the web, invoke actions in apps and retain conversations. Depending on the request or feature, processing may happen on the iPhone, through Apple’s Private Cloud Compute system, via web retrieval or through third-party visual results.

The result is a more capable assistant, but also one that asks users to make more consequential choices about access, history and data processing. Apple supplies controls for some of those choices. It does not, however, publish an exhaustive table showing how every request is routed, and its privacy assurances remain principally vendor claims rather than independently established conclusions.

Availability comes with several qualifications

Siri AI did not arrive as a universal replacement for the existing assistant. Apple describes it as a beta subject to hardware, language, regional, age and usage restrictions. On iPhone, its documentation lists the iPhone 15 Pro, iPhone 15 Pro Max and iPhone 16 models or later, running iOS 27 or later with compatible language and region settings.

The initial release was in English. Apple said French, Japanese, Korean, Portuguese and Spanish would follow in October 2026. That timetable remains a company commitment rather than an independently confirmed rollout. At launch, the beta was unavailable on iOS, iPadOS and watchOS in the European Union, unavailable in China, and not offered to users under 13.

Apple also says Siri requires an internet connection. Some server-side Apple Intelligence features, including Siri AI, have daily limits that can vary according to the feature, request complexity, demand and policy. The documents reviewed for this article give no single numerical allowance, although Apple says expanded access will eventually be sold for a fee.

For Australians, Apple’s global launch information suggests that eligible devices may be included because Australia was not listed among the regional exclusions and English was the first supported language. That remains an inference, not confirmation for every Australian Apple Account or device. Account availability, region and language settings still require checking at publication.

Personal context is the central capability

The most important change is not a more expressive voice. It is the assistant’s documented ability to find and combine information from across a person’s digital life. Apple says Siri AI can search personal context in Photos, Mail, Messages, Calendar and other app data, respond to follow-up requests and perform tasks such as drafting, sending or organising content.

This changes the nature of an assistant request. A question can depend not only on general knowledge, but also on a detail buried in a conversation, an event in a calendar or an image in a photo library. Apple also documents web retrieval for current information, allowing Siri AI to combine external material with personal context.

Apple’s guides say the assistant can interpret material already visible on the phone. They describe questions about onscreen images and PDF documents, searches and actions based on screenshots, and a Siri mode in Camera that can identify, search for and act on objects in the frame. Write with Siri can generate text, rewrite it, proofread, summarise and provide feedback, while allowing users to compare revisions and refine the output.

These abilities do not establish unrestricted access to every file or app. Third-party support depends partly on developers. Apple provides App Intents schemas and onscreen view annotations that let apps expose content and actions to Siri AI, but the existence of that framework does not mean every app supports every possible instruction. The reviewed sources contained no comprehensive matrix of compatible third-party apps and actions.

Beta performance also remains difficult to generalise. In a WIRED hands-on report, Reece Rogers said indexing took a little over a week on his test device before Siri began surfacing useful information from messages and interpreting onscreen context. That experience illustrates both the feature’s potential and the delay that may precede it. It is one reporter’s test, not a benchmark for accuracy, reliability or indexing time on other devices.

On-device does not mean local only

Discussion of Siri AI privacy can become misleading when two different processes are treated as one. Apple describes an on-device personal-context index, but that does not establish that every request and response remains on the device.

Apple says many Apple Intelligence requests run on-device. More complex work can be sent to Private Cloud Compute, its system for processing requests on servers. According to Apple, only the information needed for the request is processed there, the data is not retained or made accessible to Apple, and the server software can be inspected by researchers. Apple’s iPhone guide also describes an exportable activity report showing requests sent to Private Cloud Compute.

Those safeguards distinguish Apple’s documented design from conventional cloud collection, but they remain attributed claims. The sources reviewed for this article do not include an independent audit of the production service or verify its operation request by request.

The infrastructure has also become more complicated. In June 2026, Apple disclosed that it was extending Private Cloud Compute workloads to Google Cloud with Google and NVIDIA, and that technology behind Google’s Gemini models had contributed to the next generation of Apple Foundation Models. That disclosure is not evidence that Google or NVIDIA can read Siri requests. Apple says its Private Cloud Compute protections continue to apply, although that assurance was not independently validated in the sources reviewed here.

Web retrieval creates another route. Apple says searches and requests used to obtain current information are associated with a random identifier rather than the user’s Apple Account. The public documentation does not provide a prompt-by-prompt routing map, so users cannot determine from the wording of every request whether it will stay on the phone, use Private Cloud Compute or involve a web service.

Conversation storage is separate again. The dedicated Siri app can retain history and synchronise it through iCloud across devices using the same Apple Account. Users can search and pin those conversations. Apple provides retention choices, including options under which conversations older than 30 days or one year are permanently deleted.

A further distinction applies to the optional Improve Siri & Apple Intelligence programme. Apple’s privacy notice says that, when a user opts in, the company may store interactions including audio, transcripts, typed requests and Siri responses. The setting therefore changes the relevant data treatment and deserves separate consideration from the processing needed to answer an ordinary request.

The controls form a partial privacy dashboard

Apple does not require users to accept every contextual function as a single package. Siri settings include controls for learning and suggestions on an app-by-app basis. These are relevant when a person is comfortable allowing the assistant to use one source, such as a calendar, but not another.

Visual Intelligence has its own consequential settings. Apple says it can automatically search the web or send highlighted objects to third parties to return results. Reviewing those controls narrows when material from the screen or camera is used beyond the immediate device context.

Conversation retention can also be shortened, while the Private Cloud Compute activity report offers some visibility into off-device processing. Users can review whether they have joined Improve Siri & Apple Intelligence, rather than assuming that optional contribution is necessary for the assistant to function.

These controls do not reveal every routing decision or downstream service. They nevertheless let users reduce particular forms of access and persistence. Apple also allows Siri AI to be turned off entirely in favour of Siri Classic.

A beta defined by boundaries as well as abilities

Even some interface features have narrower compatibility than the main assistant. Apple limits Siri Expressive Voices on iPhone to the iPhone 17 Pro, iPhone 17 Pro Max, iPhone Air, iPhone 18 Pro and iPhone 18 Pro Max. That model-specific list underlines why broad claims about what Siri AI can do need to be checked against the actual device.

The beta’s practical value will also depend on how quickly personal information is indexed, which apps expose useful actions, when server limits intervene and how consistently requests are interpreted. Apple’s documentation establishes the intended functions and available controls, not uniform performance across accounts and devices.

Siri AI therefore cannot be understood through the simple claim that it knows more while everything stays local. Its defining feature is movement across boundaries: between general knowledge and private context, between the device and cloud processing, and between Apple’s software and participating apps or web services.

The meaningful adoption decision sits at those boundaries. Before the beta becomes routine, the settings for app learning, visual search, conversation retention and optional improvement data show what a user has chosen to share or preserve. The unanswered part is just as important: without a complete routing map or broad independent testing, users still have to rely partly on Apple’s description of what happens after a request leaves the screen.

The boundaries among on-device processing, cloud systems and web retrieval provide a practical reason to study networking and cloud concepts. Broaden that technical perspective through The Hack Academy’s online cybersecurity training programme, where networking and cloud security are listed subjects and learning is self-paced, with practical exercises alongside theory.