Apple may be willing to look outside its walls to get Siri up to snuff with artificial intelligence features. According to a report from Bloomberg, the company is considering using tech from Anthropic or OpenAI to right its struggles in AI.
This would be a massive change for Apple, which is currently using its in-house Apple Foundation Models to power its Apple Intelligence AI features.
The Cupertino, California-based company has reportedly spoken to the two companies about training their models so that they can be deployed on Apple’s own cloud infrastructure. Apple hasn’t decided if it’s using external models, and it reportedly still has a version of Siri using its own models in development.
A version of Siri with AI features is expected in 2026, following delays from a version that was expected in iOS 18, iPadOS 18, and macOS 15.
Apple already uses OpenAI for some Siri features, answering search queries and creating generative content. But those are effectively an extension that you have to enable, and Apple still controls its assistant.
Using a model from Anthropic or OpenAI could boost Apple into AI relevance and match what some other hardware manufacturers have done. Samsung has its own Galaxy AI features, for example, but also uses Google’s Gemini (and that’s the entire interface, not just the models). There have also been reports that Apple has considered outright purchasing Perplexity.
Apple launched the iPhone 16 last year as “built for Apple Intelligence,” but since then has appeared to be floundering in AI as it turned Siri over from AI head John Giannandrea to software engineering lead Craig Federighi and Siri boss Mike Rockwell (who previously launched the Vision Pro).
Bloomberg states that switching to an external AI model is hurting morale among the AI team at a time when some tech companies are on the hunt for new talent and are willing to pay top dollar.
Apple Intelligence and Siri don’t just affect the iPhone. They’re also on the iPad and the Mac. In losing Siri, Giannandrea also lost Apple’s robotics department. And the Siri delays also reportedly delayed a new piece of hardware, a mix between the HomePod and iPad that would work as a home assistant, similar to an Amazon Echo Show.
Apple’s own models don’t seem to be going anywhere anytime soon. At WWDC, the company said it would open its on-device models, the likes of which are used for Genmoji and Writing Tools, to developers this year, letting third-party developers build AI features using its own tools.
But those local models, while private, are slower and have proven less impressive than what can be done by competitors in the cloud. (Apple does have servers for some cloud-based requests.) Running a third-party model on Apple’s servers may give Apple a mix of the two worlds, touting user privacy while also giving it a boost to catch up in the AI race.