ChatGPT's privileged position on the iPhone may have lasted only about a year and a half.
On May 5, Bloomberg's Mark Gurman reported that Apple is preparing an Extensions system for iOS 27, iPadOS 27 and macOS 27. The idea is straightforward: users would be able to choose a default AI model in system settings, putting Claude, Gemini and ChatGPT in the same list and letting the user decide what powers Siri.
Apple is expected to unveil the changes at WWDC on June 8.
How Extensions would work
Gurman described Extensions as a way to use generative AI from installed apps on demand through Apple Intelligence features, including Siri, Writing Tools and Image Playground.
In practice, an AI app installed from the App Store could become a capability provider for Apple Intelligence. Inside Settings, under "Apple Intelligence & Siri," users would pick a default model. Once selected, Siri, Writing Tools and Image Playground would all use that model. This is not a feature-by-feature switch; it is a system-level default.
The reported list of replaceable capabilities is broad:
- Siri conversations and command understanding
- Writing Tools text generation, rewriting and summaries
- Image Playground image generation and editing
- System-level text and image processing tasks
Apple's own on-device models would still run as a fallback. But if a user chooses Claude in Settings, Claude would be the model handling a Siri mail summary, a Writing Tools rewrite or an Image Playground prompt.
The overlooked voice detail
The most interesting part of the design is audio identity.
Apple reportedly plans to assign different Siri voices to different AI providers. A Claude response would not sound the same as an answer generated by Apple's own model. Gemini would have its own voice; Anthropic would have its own voice.
That choice makes sense for both regulation and branding. Regulators in Europe and the United States are pushing for AI content to be identifiable. If users can hear that a third party is responding, Apple gets ahead of part of that compliance problem.
It also matters commercially. If users can tell which AI they are using, the distribution or placement value paid for by an AI provider becomes visible. Without that identity, Anthropic could be powering the experience while users never realize Claude is doing the work.
The signal is that Apple is not merely opening the gate reluctantly. It appears to be designing an entry point where multiple AI companies can build brand value inside the iPhone.
Who loses and who benefits
OpenAI has the most to lose.
Since 2024, Apple Intelligence has treated ChatGPT as the default external model. That gave OpenAI potential exposure across hundreds of millions of iPhones. Apple did not pay for it, but the distribution value was still enormous for an AI company chasing platform-scale reach.
With iOS 27, ChatGPT would sit beside Claude and Gemini as one option among several. Its exposure would likely fall sharply from the "default add-on" position.
Anthropic is the clearest beneficiary. Consumer distribution has long been Claude's weak spot, and Anthropic has acknowledged that ChatGPT is far ahead in consumer reach. A system-level iOS entry point would put Claude on equal footing with ChatGPT and effectively hand it one of the most valuable acquisition channels in technology.
Google's Gemini is a partial winner. Apple and Google had already discussed deeper Gemini-Siri integration, reportedly in a way that would not make the Gemini brand obvious to users. A user-visible model picker would turn that hidden presence into explicit brand exposure.
Questions Apple still has to answer
First, it is still unclear what Apple's own default model will be. Gurman's report says Apple has renewed work with Google around Gemini as a default. If Gemini is the default, OpenAI takes a second hit. If Apple's own model is the default, Google is effectively downgraded. WWDC should clarify that point.
Second, data handling remains unresolved. Mail, documents and Image Playground inputs could involve sensitive user data. Apple has long made privacy a selling point, but third-party model access implies that at least some data may leave Apple's own systems. MacRumors did not detail the privacy framework, so this is likely to be a major WWDC topic.
Third, the economics are still unknown. Claude and Gemini could pay Apple per API call, in a model that resembles App Store revenue sharing. Or Apple could charge providers for placement, closer to a premium advertising slot. Those two models would create very different cost structures for AI companies.
Apple is being pushed into this shift
Complaints about Apple Intelligence have accumulated for a year and a half. Siri still disappoints, Writing Tools are uneven, and Image Playground trails cloud models in quality. Apple's on-device models are in a much smaller class than cloud systems such as Claude Opus or Gemini 2.5 Pro.
Forcing users to rely on a model that feels several generations behind would be a poor experience. Opening the system to third parties is an admission that Apple may not catch up quickly. It is also a pivot from hardware distribution to control of AI traffic.
Apple does not need to make money from AI models directly; it makes money from iPhones. If weak AI features push users toward Android, the cost is far higher than sharing the system with Claude or Gemini. Better AI can protect the hardware business.
OpenAI's response may have to be more aggressive hardware and operating-system work of its own. Sam Altman's investment in Jony Ive's io project at least looks strategically consistent with that direction.
Sources: iOS 27 will let you choose between Gemini, Claude, CocoLoop, and more for AI features: report (9to5Mac); iOS 27 Will Let You Pick Claude or Gemini Instead of ChatGPT for Apple Intelligence (MacRumors); CocoLoop