It has been reported that Apple has designed the new Siri architecture with a private Extensions Framework that would allow users to set third-party AI models, such as Claude, as the default AI for their iPhones and Macs, based on a reverse engineer’s findings.
It is reported that the feature is compatible with iOS 27 and macOS Golden Gate, both of which are currently being made available to compatible devices. The framework includes a specific Settings section in which users can select their preferred backend model for Siri’s writing tools and image generation, Siri functioning as an intermediary that sends the commands to the chosen provider. The feature is not yet available to users.
A person known as pdfu, who is a coder and reverse engineer, showed off the feature before any official announcement was made.
How Siri Could Use Third-Party AI Models
The reverse engineer says that Apple has included two systems in order to integrate third-party AI into its hardware.
The first instance makes use of a Model Delegation API, enabling third-party models such as Claude to function as a Siri extension, just as the built-in ChatGPT extension does within Siri’s model picker.
To show this, the tipster opened the “Search or Ask” bar on a Mac, chose Claude from the “Ask” menu and asked Siri to have Claude create a reminder in the Reminders app.
The task had to be carried out by Siri since it involved access to a system app and raised potential privacy issues, after Claude had first processed and interpreted the request.
On the other hand, Claude carries out some tasks where Siri is unable to. A particular example seems to illustrate Claude producing a CSV file in response to a natural language voice command, since the Siri AI still cannot handle CSVs and some of the other file formats.
Apple’s Broader AI Model System and Availability
The second system covers a broader area and enables the inference provider within the “Model Manager Services” component of iOS 27 and macOS Golden Gate to use third-party services in place of the server-side Siri AI. However, the tipster has named GPT-5.6 as the provider, even though Apple has not confirmed the arrangement.
The person who provided the tip made it clear that the technology enables the third-party model to handle the prompt and generate the responses before passing them on to Siri, which then displays them through its own interface or speaks them aloud.
In the case of a task instead of a question, the third-party model deals with the request and passes it to Siri, which carries out the task in accordance with the user’s instructions.
Even though the model selection feature is included in iOS 27 and macOS Golden Gate, it is not currently available to users. Since the EU’s Digital Markets Act requires that built-in AI chatbots on computers give users the option of using different models, it seems likely that Apple will eventually open Siri up to third-party models. However, Apple has not yet officially confirmed the feature, the agreement with GPT-5.6, or when it will make the feature available.
I should point out that I don’t have the ability to carry out searches or access a live database, which means that I can’t verify on my own the claims made through reverse engineering, and the details are based entirely on the tipster’s demonstration and not on an official announcement from Apple.
