Apple is reportedly working on a feature called Extensions for iOS 27, iPadOS 27, and macOS 27 that would let users select the AI model powering Siri, Writing Tools, and Image Playground. Models from Google and Anthropic are currently being tested in pre-release versions of the software.
Key Takeaways
- iOS 27 would include “Extensions,” a feature for choosing third-party AI models
- Google and Anthropic are in active testing; ChatGPT’s status remains unclear
- A strategy shaped by new CEO John Ternus following Tim Cook’s departure
Extensions: what the feature actually does
According to information drawn from test versions of the software, Apple is developing a feature for iOS 27 called Extensions. Its core function: giving users the ability to choose which AI model powers the intelligent features on their device. Siri, Writing Tools, and Image Playground are the first surfaces expected to support the feature.
This represents a significant shift in philosophy for Apple. Since the launch of Apple Intelligence in 2024, the company’s approach had been to control the entire stack: the model, the inference layer, and the interface. Extensions would break from that logic by opening part of the system to third-party model providers, under Apple’s oversight.
Competitive pressure is part of the story. Apple is widely perceived as lagging in generative AI. While rivals release increasingly capable models in rapid succession, Apple has struggled to demonstrate that its integrated AI justifies the premium positioning of its hardware. Opening the platform to stronger models is a way to work around that structural weakness.
The feature would be available across iOS 27, iPadOS 27, and macOS 27, giving it a broad footprint from day one. It would cover Apple’s main device categories in a coordinated update, from iPhone to Mac to iPad, rather than rolling out piecemeal.
For end users, the practical change is significant. They would no longer be locked into the model Apple selects by default. They could choose whichever model they consider most capable, most suited to their use cases, or simply the one they trust most. That kind of user-level control does not exist today in the Apple ecosystem.
Google, Anthropic, and the uncertainty around ChatGPT
In the test versions reviewed, two third-party models are currently in active integration: those from Google and those from Anthropic. The presence of both simultaneously is notable. These are two of the most consequential AI partnerships being formed across the industry right now, and having both integrated natively into iOS 27 would signal a meaningful commitment to genuine openness.
ChatGPT’s situation is less clear. OpenAI’s model is already accessible in certain contexts on iOS, but its status within the Extensions framework has not been confirmed. The relationship between Apple and OpenAI has shifted since the early integrations announced in 2024, and it is not established whether ChatGPT will be available as an option at launch.
Partner selection likely reflects a combination of technical, commercial, and strategic considerations. Any model integrated natively into Apple’s platform must meet Apple’s requirements around user privacy, on-device performance, and content guidelines. Not every provider will qualify for native integration, which turns Extensions into a certification gateway as much as an openness initiative.
For Google and Anthropic, native placement in iOS 27 represents privileged access to hundreds of millions of active users. It is a distribution channel with no real equivalent, one that could meaningfully accelerate the adoption of their models in everyday use cases rather than purely developer contexts.
The commercial logic runs in both directions. Apple monetizes these access agreements. Google and Anthropic accept Apple’s terms in return, on data handling, integration format, and usage policies. Both sides have something to gain, which makes the arrangement durable as long as the quality bar is maintained.
Also on Horizon:
- Anthropic x SpaceX: Claude Code doubles its limits
- Google I/O 2026: Gemini is getting a new look
- Claude: how to stop wasting your tokens
John Ternus’s strategy: hardware first, open AI second
Extensions fits squarely within the vision new CEO John Ternus is building for Apple following Tim Cook’s departure. His approach rests on a clear principle: turn Apple’s existing hardware into an AI-centric experience, rather than attempting to build a model infrastructure that rivals OpenAI or Google.
That is a bet on technical humility. Apple is implicitly acknowledging that it cannot compete with AI research labs pouring tens of billions of dollars into model training. Where it has an unmatched edge is in hardware, security, and user experience design. The strategy leans into those strengths rather than fighting on ground where it is structurally disadvantaged.
Extensions makes that positioning concrete. Apple’s message shifts from “our AI is the best” to “our platform gives you access to the best AI of your choosing.” That is a significant reframing, one that turns the iPhone into a universal AI access layer rather than a proprietary AI product.
The approach is not without risk. Apple becomes dependent on the quality and reliability of partners it does not fully control. If an integrated model produces problematic outputs, the user experience damage falls on Apple’s brand, even if the technical failure belongs to the provider.
iOS 27 is expected to be announced at WWDC on June 8, 2026, with a public release in the fall. The Extensions announcement at that conference will be closely watched, notably by developers, model providers, and investors who have been waiting for Apple to demonstrate it can actually execute on AI.
Follow the story on Horizon.
Pingback: Google I/O 2026: Gemini is getting a new look - Horizon
Pingback: Anthropic x SpaceX: Claude Code doubles its limits - Horizon