macOS 27 Golden Gate Arrives With Siri AI Beta for Apple Silicon Macs
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
Golden Gate runs only on Macs with Apple Silicon, Apple's own processors. Macs with older Intel processors are not supported.
The main new feature is the Siri AI assistant. It ships in macOS 27 as a beta, available only in English at launch. A separate Siri app stores an archive of chats.
Siri AI can draw on personal context, meaning material such as your photos and emails, for search. It can also take actions inside apps, including Messages and Reminders.
There are also changes to appearance. macOS 27 uses a consistent corner shape for windows. Liquid Glass, Apple's translucent style for controls and panels, gets more uniform light effects and better contrast. A slider lets users make Liquid Glass elements more or less see-through.
Staged rollout, caps and regional split
Apple Intelligence and Siri AI across iOS 27, iPadOS 27, macOS 27, watchOS 27 and visionOS 27 require iPhone 16 models or later. Apple On Mac, Siri AI is a beta for users with a supported device set to English.
Siri AI will still be in beta when the OS 27 versions reach the public on September 14. AppleInsider Use at launch is subject to daily usage caps.
Availability differs by platform in Europe. 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
Developers had access before release. New Siri AI features have been available for developer testing across iOS 27, iPadOS 27, macOS 27 and visionOS 27. Support for watchOS 27 follows separately, in a future watchOS 27 beta.
watchOS 27 runs on recent watches only. 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 control of demand during testing. English-only support, limits to supported devices, beta status and daily caps hold down use while Apple gathers data from real use. For an assistant that reads photos and email and acts across apps, that caution is understandable. Early data on speed and reliability will lean toward English-speaking users on newer hardware.
Looking at what this means for developers, testing gets more complex. Code that calls Siri AI will need a fallback for three cases: older devices outside the support window, EU iPhones and iPads with no Siri AI at launch, and users who reach a daily cap during use. Search across personal material and actions in Messages and Reminders increase the need for careful permissions, limits on what prompts can do, and clear logs. The chat archive should be treated as stored data that users will want to search, export and delete.
Looking at what this means for fleets of Macs, the Intel cutoff cuts both ways. Apple Silicon only means one chip family to test for OS 27, one set of expectations for on-device processing, and a clearer schedule for retiring Intel machines still in service. Machines that cannot upgrade will not get the new assistant, the Liquid Glass updates or the window changes, which will bring upgrade plans forward.
In my view, the Siri app itself is the change to watch. Siri shifts from a brief voice overlay to a saved, text-based record of work. I have watched my own adult children treat chat histories as filing cabinets, returning to search them long after. If that habit carries to Siri, accurate retrieval from photos and email, and clear records of what was done in Messages and Reminders, will count more than smooth replies.
Another point worth flagging is the split in Europe. Offering Siri AI on macOS and visionOS in the EU while holding it from iOS and iPadOS creates a break for people who move between Mac and iPhone during the day. For developers with Apple Intelligence permissions, checking features by system and region will be required. The coding task is workable. The harder task is explaining to users why the assistant is present on one screen and absent on another.
Looking further ahead, these limits suggest the next steps. Caps manage computing cost. Language limits manage testing cost. Hardware limits reduce variation in on-device speed. Once those factors settle, wider language support and higher limits become practical choices rather than research questions. Past shifts from on-device to cloud-assisted computing suggest that early limits can create room for steady everyday use later.
