Siri AI in iOS 27 reportedly fails on locked devices and cloud-dependent tasks

The rollout of Siri AI in iOS 27 marks a significant architectural shift from the previous on-device assistant to a system heavily reliant on cloud infrastructure. While the new version promises richer knowledge access and onscreen awareness, the transition has introduced a dependency on network connectivity that the old Siri did not have. This change means that basic interactions are now subject to the reliability of Apple's cloud backend and the user's local network conditions, creating a new failure mode for a feature that was previously expected to work offline or with minimal latency.

A critical regression for users is the requirement for the device to be unlocked to process many requests, even when the 'allow Siri when locked' toggle is active. The material suggests this is tied to the new system's need to access a semantic index for personal context, which fails when the device is locked. This breaks established workflows for users who rely on Siri in hands-free contexts, such as driving or exercising with AirPods, where unlocking the phone is impractical or unsafe. The result is a less capable assistant in the specific scenarios where voice control is most valuable.

Performance issues are compounded by the reliance on Private Cloud Compute for much of the processing, leading to noticeable delays in response times. This is particularly evident on Apple Watch, where simple commands like starting a timer, which were previously processed almost instantly, now take many seconds to resolve. The latency is directly linked to network speed, meaning users with slower connections will experience a significantly degraded experience. This shift moves the bottleneck from local hardware to network infrastructure, making the assistant's responsiveness variable and less predictable.

Beyond performance, the new system is exhibiting indexing and feature gaps that frustrate users who expected an upgrade. Reports indicate that Siri AI often fails to find existing items like Reminders lists or HomeKit accessories, suggesting bugs in the underlying search indexing. Additionally, certain legacy features, such as querying Health app data or using the 'Use ChatGPT' prefix, are not supported in the new system. These omissions, combined with the ambiguity introduced by the new system's broader scope, mean that some previously reliable commands are now either unsupported or prone to misinterpretation.