A report from The Next Web says the iOS 27 developer beta contains a dormant Extensions framework that would let users select a third-party AI provider -- including Claude, ChatGPT, and Google Gemini -- inside Siri, complete with a settings panel and a dedicated App Store section. The feature appears disabled on Apple's backend and was not announced during the WWDC 2026 keynote. The Next Web, citing Bloomberg, reports Apple has held entitlement discussions with OpenAI, Anthropic, and Google but has not publicly committed to a multi-provider launch. Two factors appear to have delayed the announcement: Apple confirmed at WWDC that Siri AI will not ship in the EU due to ongoing Digital Markets Act negotiations, and Bloomberg has reported that OpenAI is exploring legal options over its existing Siri partnership, including a possible breach-of-contract notice.
What happened
A report from The Next Web says the iOS 27 developer beta contains an Extensions framework designed to let third-party AI providers plug into Siri, with references to ChatGPT, Anthropic's Claude, and Google Gemini. The Next Web reports the beta includes a settings panel and a dedicated App Store section for Extensions, but that the functionality appears disabled on Apple's backend. The Next Web cites Bloomberg that Apple has held entitlement discussions with OpenAI, Anthropic, and Google. Extensions for AI providers beyond ChatGPT were not announced during the WWDC 2026 keynote, even as Apple unveiled Siri AI -- a Gemini-powered rebuild of its voice assistant -- and confirmed iOS 27's general Extensions architecture for user-selectable AI features.
Technical context
The Extensions framework, as described by The Next Web, is a mechanism for routing Siri queries to a user-selected third-party LLM. Apple's rebuilt Siri AI runs on a custom Gemini model of approximately 1.2 trillion parameters (confirmed by multiple outlets), licensed from Google and routed through a three-tier architecture: on-device models handle simple tasks, Apple's Private Cloud Compute handles mid-level requests, and the heaviest queries reach Nvidia GPUs in Google Cloud, per The Next Web's WWDC coverage. How entitlement gates, sandboxing, and per-provider privacy controls would apply when multiple remote LLMs can access personal context has not been publicly defined by Apple.
Context and significance
Three factors appear to explain why Apple omitted Extensions from WWDC. First, Apple confirmed at WWDC that Siri AI will not launch in the EU at iOS 27 release, citing unresolved European Commission talks under the Digital Markets Act. A multi-provider LLM selection system carries distinct regulatory implications in that context. Second, Bloomberg has reported that OpenAI is weighing legal options over its existing Siri partnership, including a possible breach-of-contract notice, partly over concerns that ChatGPT lacks prime placement within Apple's ecosystem. Announcing a system that would formally end ChatGPT's exclusive position could complicate those negotiations. Third, earlier reporting from Mark Gurman and TechCrunch framed Extensions as the successor to the bilateral ChatGPT deal -- a "choose your own adventure" for AI models -- but no official timeline has been disclosed.
What to watch
Observers should track whether Apple enables the Extensions backend in later iOS 27 betas, any regulatory statements from the European Commission, and any public filings or statements from OpenAI. For platform and privacy teams, how Apple structures entitlements for competing LLMs inside a system assistant is a meaningful precedent for multi-model integration on a major mobile platform.
Scoring Rationale #
The story reveals Apple's intended multi-provider AI access model and the regulatory and legal factors that delayed its announcement -- context meaningful for AI/ML practitioners on Apple's platform. The 'backend disabled' claim is single-sourced from an article that could not be independently fetched; surrounding context (Extensions absent from WWDC, EU blocking, OpenAI dispute) is well-corroborated by multiple independent outlets. Score adjusted from 6.9 to 6.5.
Practice interview problems based on real data
1,500+ SQL & Python problems across 15 industry datasets — the exact type of data you work with.