Apple releases Siri AI: the assistant was rebuilt from scratch, on foundation models custom-built with Google's Gemini

Three words matter most in the Siri AI capability list: personal context, onscreen awareness and world knowledge. The official example shows the difference well. If a user asks Siri what a family member suggested as an activity during an upcoming visit, it searches the user's messages, finds that they mentioned wanting to bake a family recipe together, pulls the cooking instructions from an email the relative sent earlier, and adds the ingredients to a grocery list in Reminders. That chain involves three moves: retrieval across apps, assembly across sources, and landing as a real action. Onscreen awareness supplies the other half: while viewing a favorite team's website, a user can ask Siri to find all upcoming home games and add them directly to the calendar. World knowledge frees questions from your own data so Siri can answer about a football score, vacation planning or the best local coffee spot in a new city.

Invocation is spread wide this time, and each platform gets a different path. Apple says that in addition to saying Hey Siri, iPhone users can invoke Siri with the side button or swipe down from the Dynamic Island to start a conversation and get an in-depth answer. On iPad and Mac, Siri AI is integrated into Spotlight so users can search for answers to almost any question, and it enters systemwide context menus so users can Control-click to ask about images, files or text on screen. On Apple Vision Pro, Siri AI uses spatial computing with a 3D visualization users can place anywhere in their space, invoked by looking at it and starting to speak. There is also an all-new dedicated Siri app that uses iCloud to privately sync conversational history across a user's products, so a chat started on iPhone can continue on Mac, iPad, Apple Watch or Apple Vision Pro.

Voices and dictation got a new engine. Apple says Siri AI offers more expressive voices and a major boost in accuracy, with systemwide dictation powered by AFM Core Advanced, its most advanced on-device model yet, on products that support it. Users can customize the expressiveness and pace of Siri's voice. On dictation, Apple says it captures what users say as polished text, automatically handling capitalization, punctuation and formatting as they speak. Visual Intelligence also expanded from iPhone to iPad, Mac and Apple Vision Pro: on iPhone its multimodal capabilities are integrated into the Camera app with a brand-new Siri mode, so looking at a music festival poster you can ask for set times and have Siri create multiple calendar events for the weekend; on iPad it is built into the screenshot experience; on Mac a dedicated keyboard shortcut lets you select something on screen and type directly to Siri.

Writing capabilities are far more specific than before. Apple says Siri now offers integrated Writing Tools more powerful than ever, so users can write with Siri AI virtually anywhere they type: describe what you need and Siri can generate a draft from scratch, and if you want to refine it, describe the change and Siri updates it. When writing in Mail and Messages, Siri can reflect how users usually communicate with each recipient, including the punctuation and tone they typically use. Apple Intelligence now also proofreads automatically as you type across the system, including within most third-party apps. Support for outside apps is expanding too: users can already use Siri to text a friend in WhatsApp or play a book on Audible, and Apple says that soon they will be able to ask Siri AI to draft an email in Microsoft Outlook, search for a homework assignment in Notability, or add a restaurant recommendation to Tripsy.

Architecture and privacy are the heaviest parts of the post. Apple states that these new capabilities are powered by the next generation of Apple Foundation Models, custom-built in collaboration with Google and its Gemini models for deeply integrated Apple Intelligence experiences, running on device and on servers using Private Cloud Compute. Apple's claim is that when Private Cloud Compute handles users' requests, their personal data is not stored nor made accessible to Apple or anyone else, and outside experts can continue to verify this privacy promise at any time. Siri itself uses the system orchestrator to tap core capabilities like the Spotlight index and App Toolbox, which work entirely on device to keep users in control of their data. The post's closing claim is unambiguous: Siri remains the world's most private digital assistant.

Another batch of capabilities lands later this year. Apple says Apple Watch Series 12 and Apple Watch Ultra 4 bring a new class of Apple Intelligence features grouped as Audio Intelligence, designed to help users stay aware of the world around them, catch what they missed and remember key moments. The privacy design is spelled out: these features do not create or store audio recordings, and the raw audio used for processing is completely inaccessible to the operating systems, apps, the user or Apple, because a hardware-isolated Secure Exclave in the S11 chip processes audio in complete isolation and then immediately deletes it. Features are opted into individually, and Siri Recap produces a brief high-level summary rather than a transcript, designed to exclude sensitive information such as financial information or government-assigned identifiers. Live Rewind plays an audible chime while in use, even if the watch is on silent, and shows a full-display animation and microphone indicator. The Health app also gets a new design, turning metrics into personalized guidance with Apple Intelligence at its core and a summary on the new Insights tab. Separately, 9to5Mac notes Siri AI is available only on newer devices, with some voice-customization abilities limited to specific hardware.

One note on how to read a launch like this. Set alongside this September's wave of AI news, Apple's route here has two checkable traits. First, the capabilities are built around data already on the device: personal context comes from your messages, emails and photos, and onscreen awareness comes from the interface in front of you, and neither requires moving that data elsewhere. Second, Apple volunteered the partner in its own post: the foundation models are custom-built in collaboration with Google and its Gemini models. For readers used to equating Apple with wholly in-house work, that line is a fact worth updating. As for how stable the beta proves and when third-party apps genuinely plug in, the official post offers no timeline, and only real usage records can answer those questions.