Apple’s mobile operating system is quietly preparing for a massive shift in how it handles artificial intelligence. According to hidden code frameworks uncovered by a researcher in the release candidate of macOS Golden Gate and iOS 27, Apple has engineered its new Siri architecture to support deep integration with third-party AI models like Anthropic's Claude and OpenAI's ChatGPT.
The discovery, shared on X by code sleuth "pdfu," reveals that users may soon be able to bypass Apple's native Siri intelligence in favor of more capable third-party models, without losing access to native system tools. For small and mid-sized businesses relying on AI to automate their operations, this bridges a major gap between mobile hardware and the customized AI models companies actually use to get work done.
Under the Hood: Two Paths to Integration
The private frameworks reveal two distinct ways Apple is preparing to open its ecosystem to outside AI providers.
The first is a mechanism called "Model Delegation." This allows third-party models like Claude to act as an extension of Siri, similar to how Apple’s built-in ChatGPT integration currently functions. In a demonstrated example, a macOS user selects Claude from an "Ask..." contextual menu and requests a reminder. Claude processes the natural language, but seamlessly hands the task back to Siri to actually execute the creation of the reminder in Apple's native app. It also allows the third-party model to handle tasks Siri cannot do natively, such as generating a CSV file.
The second protocol, found within "Model Manager Services," is far more comprehensive. It includes an inference provider feature that can completely replace Apple’s server-side Siri model with a third-party alternative like GPT-5.6 Terra.
In this deeper integration, the third-party model receives Apple's native Siri planner prompts and tool definitions. This means ChatGPT can be asked to search for emails on a specific topic, summarize the action points, and independently draft and send a message to a saved contact via the native Messages app—all while outputting the final response through Siri's familiar voice and interface.
Why the Walled Garden is Opening
Apple's sudden embrace of interoperability is likely driven by regulatory pressure. The European Union's Digital Markets Act (DMA) mandates that gatekeeper companies like Apple provide third parties with effective access to the hardware and software features available to their own native services. The European Commission has explicitly stated that this principle extends to virtual assistants like Siri.
While the public-facing features of the macOS 27 Golden Gate Release Candidate only display the ChatGPT extension for now—and the broader model delegation entitlements remain locked away from developers and users—the underlying code proves that Apple has fundamentally re-architected its system for widespread model interoperability.
Concrete Takeaways for SMB Automation
For small and mid-sized businesses building out their automation roadmaps, Apple’s architecture shift removes a major operational bottleneck.
Currently, integrating AI into mobile field operations or remote workflows requires a patchwork of dedicated third-party apps, custom web portals, or clunky API workarounds. Employees have to leave their native mobile workflows to access a company's preferred AI tools.
By allowing models like ChatGPT or Claude to natively ingest Apple's tool definitions and execute system actions, the device itself becomes a unified automation hub. Here is what this means practically for business operations:
- Frictionless Mobile Workflows: Field technicians or sales reps could soon use Siri's voice interface to query a customized Claude or ChatGPT model, asking it to pull customer data from an email, format it, and push it to a CRM via a shortcut, all hands-free.
- Unified Tool Calling: Because Apple is providing native tool definitions to the third-party models, businesses will not have to build redundant mobile interfaces to trigger internal system actions. The third-party LLM can securely interact with native calendar, mail, and messaging apps right out of the box.
- Model Agnosticism: Operations teams won't be locked into whatever AI Apple decides to build. If Anthropic releases a model that is vastly superior at analyzing complex logistical data or coding, businesses can route their employee hardware to use that specific model without sacrificing system-level integrations.
Apple has not yet announced a timeline for opening these delegation entitlements to third-party developers, but the foundation is fully built. Companies mapping out their AI operational strategy should start planning for a near future where their preferred models can directly drive the mobile devices their teams use every day.