Private Apple Hooks Let Third-Party AI Models Work Through Siri

Researchers analysing iOS 27 and macOS 27 code say Apple has included private interfaces that can support third-party Siri AI extensions, although these hooks are not available as public APIs. A developer demonstration reportedly used the interfaces to run Anthropic’s Claude from Siri’s AI query menu through the model delegation API in App Intents.

This could extend Siri’s capabilities, such as generating CSV files, while allowing requests requiring system access—such as creating calendar reminders—to be handed back to Apple’s native Siri services.

A second demonstration reportedly replaced Siri’s server-side model with “GPT-5.6 Terra” through the Inference Providing protocol in Model Manager Services. In this design, the external model receives Apple’s planner prompts and tool definitions, can request system actions, and receives the resulting information, including personal data, before returning a response through Siri’s existing interface and voice output.

The article presents these capabilities as developer testing rather than an Apple-supported feature, and does not establish that Apple has officially enabled or endorsed them.

The main privacy concern is that using third-party models may send personal information to the companies operating those services. The report cautions that sensitive requests could therefore create a risk of data leakage. No CVE, confirmed abuse or mitigation is identified in the supplied material; users should treat unofficial private-hook experiments as unsupported and consider the data-handling policies of any external AI service involved.