For the first time, the model behind your phone’s AI is going to be a Settings toggle. According to Bloomberg’s May 5 report from Mark Gurman, iOS 27, iPadOS 27, and macOS 27 will ship a feature called “Extensions” — a system-level switch that lets users replace ChatGPT with Claude, Gemini, or Grok across Writing Tools, Image Playground, and the long-tail Siri queries Apple’s own models can’t handle. Apple is expected to announce it at WWDC in June, with public release this September.

This is the part that’s actually new. Default browsers and default mail apps were already a thing. Default AI provider — at the OS level, across every system surface that calls a model — is the first time a major platform has framed competing AI labs as interchangeable utilities rather than separate apps. For anyone already paying for Claude Pro, Gemini Advanced, or a Grok subscription, the ROI math on those plans is about to change. For IT and security teams, it raises a question nobody has answered yet: can MDM lock this down on a corporate-enrolled iPhone?

Quick Summary: What’s Coming in iOS 27

Detail	Info
Feature name	Extensions
Where it lives	Settings, system-wide preference
Platforms	iOS 27, iPadOS 27, macOS 27
Surfaces it controls	Writing Tools, Image Playground, Siri model handoff
Confirmed test partners	Anthropic (Claude), Google (Gemini)
Likely launch lineup	Claude, Gemini, Grok, ChatGPT
Announcement	WWDC, June 2026
Public release	September 2026 (alongside iPhone 18)
Primary source	Bloomberg / Mark Gurman, May 5, 2026

Bottom line: Apple just turned the default AI model into a system preference. That changes how you should think about your AI subscriptions, how IT should think about MDM policy, and how the frontier labs should think about their iPhone strategy. The September release is when it ships. The procurement conversations should already be starting.

---

What Actually Happened

Bloomberg’s Mark Gurman published the original report on May 5, sourced from people familiar with Apple’s plans. Per Gurman, the feature is internally referred to as Extensions and will appear in Settings as a system-level option (distinct from the per-app choices Apple has shipped before for browsers, mail clients, and translation). Coverage from TechCrunch, MacRumors, and AppleInsider followed within hours, all reading the same way: an extension of last year’s ChatGPT integration into a true multi-model default, set once and respected everywhere the OS calls a frontier model.

The surfaces it covers are the ones Apple Intelligence already exposes. Writing Tools: proofreading, rewriting, summarization inside any text field. Image Playground (the on-device image generation feature). And the Siri “world knowledge” handoff, where Siri kicks a query upstream to a frontier model when its own foundation models can’t answer. Today, that upstream is ChatGPT by default. In iOS 27, it’s whatever the user picks.

Per Gurman’s reporting, Apple has already conducted test integrations with Anthropic and Google. xAI’s Grok is the likely third launch partner; ChatGPT remains an option but loses its de-facto default position. The architecture inherits the privacy envelope Apple built for the Gemini-powered Siri overhaul in iOS 26.4 — queries route through Apple’s Private Cloud Compute layer, the third-party model processes the prompt, and Apple controls the data boundary so the model provider doesn’t see persistent user state.

Public release is timed to September 2026, alongside the iPhone 18 lineup. WWDC in June is where Apple will demo it.

Why a System-Level Default Is Different

The cynical read is that this is just default browser, but for AI. That read misses the point.

A default browser swap changes which app launches when you tap a link. A default AI swap changes which model writes a paragraph in Notes, generates an image from a prompt in Messages, summarizes a webpage in Safari, and answers a Siri question that Siri can’t handle on its own. It’s not one app handing off to another app. It’s the operating system itself routing every “this needs an LLM” decision through a different vendor’s model.

That makes the choice stickier than it looks. Three reasons.

The surfaces are invisible. Most users will never think “I want to call Claude right now.” They’ll hit the rewrite button in Mail. They’ll ask Siri something. They’ll generate an image. The model behind those actions is now a setting, not a launch decision. Once a user picks Claude, every text field in the OS becomes Claude. The friction to switch is zero, but so is the awareness that a choice is even being made.

The subscriptions stop being app-bound. Today, paying for Claude Pro or Gemini Advanced gets you a better experience inside those companies’ apps. iOS 27 Extensions changes that — your existing subscription now powers your phone’s writing tools, your image generation, and the long-tail Siri queries. The pro tier of a third-party model isn’t a separate destination anymore. It’s the engine.

The platform owners get a new revenue lever. Bloomberg’s reporting flagged the same unresolved question: how does Apple get paid? The current ChatGPT integration is reportedly an awareness deal, not a revenue share. Multiple competing providers, each wanting to be the default, is exactly the leverage Apple has used before to extract App Store-style economics. The September release won’t ship without an answer, and the answer will affect pro-tier pricing across all four providers.

The Subscription Math Just Changed

If you’re already paying for one of these, run the numbers again.

Claude Pro is $20 per month. Gemini Advanced is bundled into Google One AI Premium at $19.99 per month. Grok’s premium SuperGrok tier is $30 per month. ChatGPT Plus is $20 per month. Until iOS 27 ships, those subscriptions are mostly app-bound — you’re paying for a better experience inside the provider’s own client.

After iOS 27, the same subscription powers your phone’s system-level AI. That’s a meaningful change in value-per-dollar. The pro tier of whichever model you pick becomes the engine behind:

• Every rewrite, summarize, or proofread action across iOS apps
• Every Image Playground generation
• Every Siri query that requires a frontier model
• (Reportedly) higher rate limits and access to the provider’s most capable models for those system surfaces

That math probably tips the choice for users currently paying for two providers. If you have ChatGPT Plus and Claude Pro and you only need one as your system default, the cost-benefit of dropping the duplicate gets sharper. Conversely, if you’ve been on the free tier of one provider, the iOS 27 launch is the moment the upgrade pitch lands across every text field on your phone. That’s the side of the trade Anthropic, Google, and xAI are going to push hardest at WWDC.

For a deeper read on how the four models actually compare on quality, the 2026 head-to-head between Claude, ChatGPT, and Gemini is the work to look at before September. The Grok 4.20 review covers where xAI’s model is genuinely competitive and where it isn’t. iOS 27 makes the choice load-bearing in a way it wasn’t before.

The Enterprise Question Nobody Has Answered

Here’s the part that should be on the agenda of every enterprise mobility team this quarter, even though nobody is talking about it yet.

Corporate-enrolled iPhones are managed through MDM tools — Microsoft Intune, Jamf, VMware Workspace ONE, and the long tail. Those tools enforce policy on a wide range of iOS settings: which apps install, which networks connect, which data flows out, which features turn on. The question Apple has not yet answered is whether iOS 27 Extensions falls inside or outside that policy surface.

Three scenarios, and they have very different implications.

Scenario one: MDM can lock the choice. IT sets the default — say, Claude or ChatGPT under enterprise contract — and users can’t change it on a managed device. This is the cleanest outcome for enterprise security, and the most likely one for Apple to ship given how every other system-level data-flow decision has been MDM-controllable since iOS 13.

Scenario two: MDM can enumerate but not enforce. IT can see which Extension is selected on each device and report on it for audit purposes, but the user makes the final call. This is the messy middle and the worst outcome for compliance teams in regulated industries — visibility without control means policy that exists on paper but doesn’t survive contact with users.

Scenario three: MDM can’t see or set it. Apple treats Extensions as a personal-preference setting (like default browser was originally) with no managed equivalent. This is the worst outcome for enterprise. A finance employee on a corporate iPhone could route every Writing Tools query through Grok, against company policy, and IT has no visibility.

Apple has not committed to which of the three it will ship. The MDM payload reference in the iOS developer documentation won’t update with Extensions-specific keys until the WWDC beta drops in June. That’s the document IT should bookmark.

The question for now is which providers your existing enterprise AI contracts cover. If you’re a Microsoft 365 enterprise customer with Copilot, an Agent 365 governance posture, and a separate Anthropic contract for Claude through Bedrock, the iOS 27 Extension you’d want as the corporate default is whichever provider already sits inside your data-handling agreements. If your security team has been quietly nervous about the multicloud sprawl of agent and model contracts, iOS 27 Extensions is the next surface where that sprawl shows up — at the personal device level, where contract enforcement is hardest.

What Apple Has to Decide Before September

The Bloomberg story is the announcement that the feature is coming. The harder questions are the ones Apple has to settle in the four months between now and ship.

Model certification. Apple has historically gated which providers can plug into iOS at the system level — App Store rules, but for OS-level integrations. Will Extensions require a certification process for the model behind them? Will it impose latency, safety, or content-filtering thresholds? A certification bar high enough to lock out smaller labs would make Extensions a feature for the top three or four providers only. A bar low enough to admit the open ecosystem changes the dynamics for open-source models like the Llama 4 family on iPhone — eventually, even if not at launch.

Revenue share. App Store-style economics on AI subscriptions would be a significant new revenue stream for Apple and a margin hit for every provider. A pure default-app model with no rev share is the friendlier outcome for the labs. The middle ground — Apple takes a referral fee on subscriptions originated through the iOS upsell flow — is the most likely shape.

Pro-tier pricing alignment. Each provider currently charges different amounts for pro tiers with different feature sets. iOS 27 Extensions creates pressure toward a comparable “iOS Extension Pro” SKU at a comparable price point — otherwise users will flatly compare Claude Pro at $20 against SuperGrok at $30 and pick the cheaper engine for the same surface. Expect pricing changes from at least two of the four providers between June and September.

MDM payload schema. As covered above, the policy surface determines whether this feature is deployable in regulated industries on day one or whether it’s a personal-device-only feature for the first year. Apple tends to ship enterprise-grade controls late, then add them in point releases. The October iOS 27.1 update is the realistic window for the first true managed configuration.

Privacy disclosures. Each provider’s Extension will ship with its own data-handling document. Apple will need a unified disclosure surface so users see, in plain language, what each model can and can’t do with their data — particularly given that the routing through Private Cloud Compute means Apple is part of the data path even when the inference happens at a third party.

What This Means for the Frontier Labs

Every frontier lab now has an iPhone strategy whether it wanted one or not.

For Anthropic, Extensions is the closest thing to a consumer distribution channel Claude has ever had. Anthropic’s commercial momentum has been driven by enterprise and developer adoption, not by the consumer app. iPhone-level default-AI status changes that overnight. The strategic risk: Claude’s brand is built on calm, controlled, enterprise-grade behavior. The system surfaces in iOS 27 — quick rewrites, image generation, casual Siri queries — are not where Claude’s strengths shine brightest. Anthropic will need to tune for the surface or accept that consumer benchmarks will read differently than enterprise benchmarks.

For Google, Extensions is the validation of a strategy already in motion. Google has been quietly building Gemini into the iPhone experience for a year — through the iOS 26.4 Siri overhaul and the deepening Apple-Google partnership. iOS 27 Extensions makes Gemini a peer, not a backend. The competitive question is whether Apple will let Google market Gemini directly to iPhone users or whether Apple will broker that relationship.

For xAI, Extensions is a credibility moment more than a revenue moment. Grok is the youngest of the four launch options, with the smallest user base outside X. Inclusion as a system-level default normalizes Grok as a peer of the other three labs. That’s a brand outcome xAI would have paid hard for. The substantive question is whether Grok’s content posture — looser safety, more willingness to engage edgy queries — survives an Apple system-level integration where the surfaces include schoolwork, business writing, and family communications.

For OpenAI, Extensions is a downgrade. ChatGPT was the de-facto default; it now becomes one of four. The current Apple-OpenAI integration is reportedly an awareness deal that pays OpenAI nothing direct. Losing the de-facto default status means OpenAI’s iPhone presence is now competitive, not embedded. Combined with OpenAI ending Azure exclusivity to ship through AWS Bedrock, the company’s distribution story is shifting from “default everywhere” to “available everywhere” — which is a different position with different economics.

Our Take

This is the kind of move that looks small at the announcement and reframes the category eighteen months later.

The right way to read iOS 27 Extensions is as Apple commoditizing the model layer on its own platform. Apple Intelligence’s first year was rocky — the on-device foundation models couldn’t do enough, the ChatGPT handoff was a stopgap, and the Gemini deal that powered Siri in iOS 26.4 was a tacit admission that Apple wasn’t going to win on raw model capability. Extensions takes that admission and turns it into a strategy. If Apple can’t beat Claude or Gemini at frontier reasoning, it can be the platform that lets users pick whichever frontier model is best, and capture the relationship at the OS layer instead of the model layer.

That’s a defensible position. It’s the same position the iPhone has held in maps (defaulted to Apple Maps, but you can pick Google Maps or Waze), browsers (default Safari, swappable), and translation. The model becomes a utility; the platform stays valuable.

For the frontier labs, the implication is harder. Being a system-level default on a billion iPhones is a once-in-a-generation distribution opportunity — and a serious pricing risk. If Apple sets the rev-share economics, the labs are negotiating against the same App Store pressure that compressed every other developer’s margins for the last fifteen years. If Apple doesn’t, the labs are getting near-free distribution at the cost of brand differentiation. Either outcome reshapes the consumer AI economics, and the WWDC keynote in June is when the shape becomes clear.

For users, the practical advice is small for now. Don’t switch subscriptions in May or June based on the announcement. Wait until WWDC reveals the actual integration and the certified launch lineup. By August, the picture will be clear enough to make a real choice — and the September release is when the choice starts to matter daily.

For enterprise IT, the practical advice is larger. Add Extensions to your iOS 27 readiness review now. Push your MDM vendor for a position on managed configuration before the WWDC beta. Map your existing AI vendor contracts to the four likely Extension providers and identify which one fits your data-handling agreements without renegotiation. The companies that have answers ready when Apple ships in September will be the ones who don’t end up in a ninety-day scramble in October.

What to Watch Between Now and September

WWDC keynote in June. The announcement that the feature exists is from Bloomberg. The official confirmation, the launch lineup, and the integration architecture come from the keynote. Watch specifically for whether Apple announces the rev-share model, the MDM controls, and the data-handling disclosure structure on stage.

Beta release notes for iOS 27 developer beta 1. The MDM payload schema for Extensions will appear in the developer beta release notes if Apple has decided to ship enterprise controls at launch. Their absence in beta 1 is the signal that managed configuration is a 27.1 or 27.2 feature.

Provider pricing adjustments. Watch Anthropic, Google, and xAI for pricing or feature changes between June and August. If any of them re-tier their pro plans to align with iOS 27 Extension positioning, that’s the strongest signal that Apple’s rev-share negotiation has settled.

Enterprise vendor commitments. Microsoft’s Agent 365 GA showed how fast enterprise AI vendors will move to slot themselves into a new control plane. Expect every MDM vendor — Intune, Jamf, Workspace ONE — to publish iOS 27 readiness positions within a week of the WWDC keynote.

The fifth provider question. If the launch lineup is Claude, Gemini, Grok, and ChatGPT, Meta’s Llama-based assistant is the obvious absence. Whether Apple admits a fifth provider at launch — or in 27.1, or in iOS 28 — will tell us how the certification bar is structured and how aggressively Apple wants to commoditize the model layer.

Frequently Asked Questions

What is iOS 27 Extensions?

Extensions is a new system-level setting in iOS 27, iPadOS 27, and macOS 27 that lets users pick which AI model — Claude, Gemini, Grok, or ChatGPT — powers Apple’s on-device AI surfaces. Per Bloomberg’s reporting from Mark Gurman on May 5, 2026, the feature replaces the current ChatGPT-by-default integration with a user-selectable choice across Writing Tools, Image Playground, and the Siri model handoff for queries Apple’s own foundation models can’t answer.

When does iOS 27 ship?

Apple is expected to announce iOS 27 at WWDC in June 2026 and ship the public release in September 2026 alongside the iPhone 18 lineup. The Extensions feature is reported to be part of the September launch, not a later point-release add. The developer beta is typically available within hours of the WWDC keynote.

Which AI models will iOS 27 Extensions support?

Per Bloomberg, Apple has already conducted test integrations with Anthropic (Claude) and Google (Gemini). xAI’s Grok is the likely third launch option, with ChatGPT remaining as a fourth. The official certified lineup will be confirmed at WWDC. Other providers — including Meta’s Llama-based assistants — are not currently confirmed for the launch lineup.

Does my Claude Pro or Gemini Advanced subscription work with iOS 27?

The expectation, based on how the current ChatGPT integration is structured, is that paid subscriptions to the supported providers will unlock higher-tier capabilities and rate limits when that provider is selected as the iOS 27 Extension. Pricing alignment across providers is one of the unresolved questions Apple and the labs need to settle before September. Free-tier access is expected as the baseline for any selected Extension.

Can corporate IT lock down which AI Extension my iPhone uses?

Apple has not yet confirmed whether iOS 27 Extensions will be a manageable setting through MDM tools like Microsoft Intune or Jamf. The MDM payload schema for the feature will appear in the iOS 27 developer documentation after the WWDC beta release in June 2026. Enterprise IT teams should treat this as an open question and plan readiness reviews accordingly. For broader context on how enterprise AI policy is evolving, the Microsoft Agent 365 governance post covers the parallel control-plane work happening on the Microsoft side.

How is this different from the current ChatGPT integration in iOS 26?

The current iOS 26 integration uses ChatGPT as the de-facto handoff for Siri queries that exceed Apple’s on-device foundation models, with no user choice of model. iOS 27 Extensions promotes that handoff to a system-level setting, lets the user pick the provider, and extends the routing to Writing Tools and Image Playground in addition to Siri. The architectural difference is that iOS 27 treats the third-party AI as a peer-replaceable utility, where iOS 26 treated ChatGPT as a single contracted partner.

Does iOS 27 Extensions affect Apple Intelligence’s on-device models?

No. Apple’s own on-device foundation models continue to handle personal-context tasks and short-form requests where privacy and latency favor local execution. Extensions only affects the upstream handoff for queries that exceed local model capability. The architecture inherits the Private Cloud Compute privacy envelope that powered the Gemini-backed Siri overhaul in iOS 26.4 — third-party models process the prompt without persistent access to user data.

Will iOS 27 Extensions cost extra?

Apple has not announced a paid tier for Extensions itself. The cost variable is the chosen provider’s subscription. Whether Apple takes a revenue share on subscriptions originated through the iOS upsell flow is one of the open questions, per Bloomberg’s reporting. The most likely shape is a referral fee on new subscriptions signed up through iOS, similar to the App Store model.

How does this affect OpenAI’s iPhone position?

OpenAI loses its de-facto default status. ChatGPT becomes one of four selectable Extensions rather than the contracted handoff. Combined with OpenAI ending Azure exclusivity to ship through AWS Bedrock, the company’s distribution position is shifting from “default everywhere” to “available everywhere” — a different competitive posture with different economics.

Should I switch AI subscriptions in advance of iOS 27?

No. Wait for WWDC in June for the confirmed launch lineup, the integration architecture, and any provider pricing adjustments. By August, the picture will be clear enough to make an informed switch. The September public release is when the choice starts to affect daily use; switching earlier is buying on a roadmap rather than on shipped behavior.

What should enterprise IT do right now?

Start the iOS 27 readiness review that includes Extensions. Map your existing AI vendor contracts to the four likely launch providers (Claude, Gemini, Grok, ChatGPT) and identify which one fits your data-handling agreements without renegotiation. Push your MDM vendor for a public position on managed configuration before the WWDC beta drops. Add iOS 27 Extensions to your enterprise AI deployment risk register so the conversation is happening before September, not after.

---

Last updated: May 6, 2026. Sources: Bloomberg — Apple Plans Letting Users Pick Claude, Gemini, Grok in iOS 27 · TechCrunch — Apple plans to make iOS 27 a Choose Your Own Adventure of AI models · MacRumors — iOS 27 Will Let You Pick Claude or Gemini Instead of ChatGPT for Apple Intelligence · AppleInsider — iPhone Users Will Get to Select a Preferred AI Model in iOS 27.

Related reading: Siri iOS 26.4 Review: Apple’s Gemini-Powered AI Overhaul · Claude vs ChatGPT vs Gemini 2026 · Grok 4.20 Review · Microsoft Agent 365 GA: What Enterprise Buyers Need · OpenAI Ends Azure Exclusivity: AWS Gets GPT-5.5 · Anthropic vs OpenAI 2026 · Enterprise AI Deployment Guide