macOS 27 Golden Gate Ships With Siri AI in Beta
Apple released macOS 27 Golden Gate on September 14. The Verge The release arrives alongside iOS 27, iPadOS 27 and watchOS 27, all slated for the same date. Engadget
Compatibility is Apple Silicon only. Golden Gate runs on all Macs with Apple Silicon. Intel Macs are not supported.
The headline workload is the new Siri AI assistant. It ships in macOS 27 as a beta, available only in English at launch. A dedicated Siri app keeps an archive of chats.
Functionally, Siri AI reaches into personal context and system actions. It can search personal context such as photos and emails. It can take actions in apps including Messages and Reminders.
Interface changes are narrow but precise. macOS 27 enforces consistent window corner radii. Liquid Glass gets uniform refraction and improved contrast. A slider controls opacity for Liquid Glass elements, making them more or less opaque on user preference.
Staged rollout, caps and regional split
Device gating is tight. Apple Intelligence and Siri AI across iOS 27, iPadOS 27, macOS 27, watchOS 27 and visionOS 27 are available on iPhone 16 models or later. Apple On Mac, Apple positions Siri AI as a beta for users with a supported device set to English.
Siri AI will remain in beta when the OS 27 versions reach the public on September 14. AppleInsider The public debut is subject to daily usage caps.
Regional availability is split by platform. Apple will not ship Siri AI in the European Union with the release of iOS 27 and iPadOS 27 due to the Digital Markets Act. Siri AI will be available to EU users on macOS 27 and visionOS 27. Apple
Developer access preceded the public release. New Siri AI features have been available for developer testing across iOS 27, iPadOS 27, macOS 27 and visionOS 27. Availability for watchOS 27 follows separately, in a future watchOS 27 beta.
watchOS 27 itself is limited to recent hardware. It will be available for Apple Watch Series 9 or later, Apple Watch SE 3, and Apple Watch Ultra 2 or later. Apple
What matters for builders and fleets
The broader context here is capacity management disguised as a beta. English-only, supported-device gating, beta status and daily usage caps together throttle demand while Apple collects production telemetry. That is a rational way to ship an assistant that touches photos, email and cross-app actions. It also means early performance and reliability data will be skewed toward U.S. English power users on newer hardware.
Looking at what this means for developers, the testing matrix just got more fragmented. Code paths that invoke Siri AI will need fallback behavior for three distinct cases: older devices outside the support window, EU iPhones and iPads with no Siri AI at launch, and users who hit a daily cap mid-session. Personal-context search and in-app actions in Messages and Reminders raise the stakes for permission handling, prompt scoping and audit logs. Builders should treat the chat archive not as UI chrome but as persistent state users will expect to search, export and delete.
Looking at what this means for fleets, the Intel cutoff simplifies the picture. Apple Silicon only means one architecture to validate for OS 27, one set of performance assumptions for local inference, and a cleaner retirement timeline for Intel hardware still in service. The cost is a hard fork. Machines that cannot move cannot get the new assistant, the Liquid Glass refinements or the consistency fixes, which will push refresh planning forward.
In this author's view, the persistent Siri app is the more consequential change. Moving Siri from an ephemeral voice overlay to an archived, text-addressable app reframes it as a work record. I have watched my own adult children treat chat histories as de facto file systems, searchable long after creation. If that habit transfers to Siri, accuracy of retrieval from photos and email, and clarity about what was acted on in Messages and Reminders, will matter more than fluency of responses.
Worth flagging is the EU asymmetry. Shipping Siri AI on macOS and visionOS in the EU while withholding it on iOS and iPadOS creates a cross-device discontinuity for users who move between Mac and iPhone during the day. For developers with Apple Intelligence entitlements, feature detection by OS and region becomes mandatory, not defensive coding. The technical burden is manageable. The product burden is explaining to users why an assistant exists on one screen and not another.
Looking at what this means longer term, the constraints point to what comes next. Caps control inference cost. Language limits control model validation cost. Hardware limits control on-device performance variance. Once those variables stabilize, broader language coverage and higher limits become operational decisions rather than research problems. The optimistic read, and the one borne out by prior platform shifts from local to cloud-assisted compute, is that staged scarcity now buys headroom for dependable, everyday use later.
