Siri Model Delegation: Code Points to Claude, ChatGPT

Code found in Apple's private internal frameworks points to a feature called Model Delegation that would let Siri hand a query to Claude or ChatGPT instead of Apple's own models.

This is a code-level finding reported by others, not an Apple announcement. Nothing here has been confirmed by Apple, and features discovered in internal frameworks routinely ship late or not at all.

What the report describes

The claim originates with MacRumors, in a story dated September 14, 2026, and was summarized in a dev.to write-up of the same finding. According to that summary, the code "discovered in private internal frameworks reveals an upcoming architecture feature called Model Delegation, built straight into the core operating system."

Two specifics are worth separating from the framing around them.

• Where it sits. The report describes Model Delegation as built into the operating system rather than layered on top of it — the write-up calls it "not a clunky shortcut, but a deep platform integration."
• How a user would reach it. The same report says users can switch models directly from the system search bar and delegate complex tasks to Claude.

The described behavior is that Siri queries can be offloaded to Claude alongside OpenAI's ChatGPT, rather than being confined to Apple Foundation models.

Why this one is worth tracking

What the report describes has a specific shape worth noting: a named delegation layer sitting in the operating system, a user-facing model switcher in the search bar, and two third-party vendors named rather than one.

If it ships as described, the practical change for users is that the quality of a Siri answer stops being a fixed property of the assistant and becomes something you choose per task. For developers, an OS-level delegation layer is the more interesting half, because it implies a routing interface that exists whether or not third parties ever get to plug into it.

What is not established

Everything about timing, availability and pricing. The report does not establish which OS release would carry the feature, which regions or languages it would cover, whether a paid Claude or ChatGPT subscription would be required, or whether the model list would stay limited to those two.

It also does not establish that the feature will ship at all. Code in an internal framework shows that something was built and tested, not that it was approved for release.

What to watch

The signal to wait for is the first beta seed in which the described model switcher appears in a shipping build rather than in decompiled framework strings. Until then, treat Model Delegation as a credible report about work in progress — and treat any spec sheet, release date or price attached to it as invention, because none of those appear in the reporting.