Siri U-turn: Apple reportedly flirting with Anthropic's AI model

Apple is reportedly considering a U-turn on its AI plans for voice assistant Siri. According to a new report, the iPhone manufacturer is considering using Claude from Anthropic or ChatGPT from OpenAI instead of its own AI model. Claude has emerged as the internal favorite, the report continues. However, Apple wants to operate a customized version of the model on its own cloud servers.

According to the US financial news agency Bloomberg, the considerations are said to be related to personnel changes that have been made in the area of responsibility for Apple's AI division. John Giannandrea, who moved from Google to Apple, reportedly lost his former responsibility for Apple Intelligence. He has been replaced by software boss Craig Federighi, who is now responsible for AI projects in general, and Mike Rockwell, who is in charge of Siri in particular. Rockwell was previously responsible for the development of Vision Pro.

Apple sticks with its own servers

Rockwell had various models (Claude, ChatGPT and Google Gemini) tested. The test series also examined whether the models could be operated on Apple's customized AI servers. The Private Cloud Compute is characterized by the fact that it should not have a "memory" in order to ensure greater data protection for users. It is not clear from the report to what extent Apple can guarantee the previous standard of Apple Intelligence with a third-party model. In any case, Apple has started talks with Anthropic. These are at an early stage.

Apple's possible decision to abandon its own AI model for Siri is linked to the delays in the restructuring of the voice assistant. Apple promised a significantly better Siri at the WWDC developer conference in 2024. This year, it was then announced that completion would be delayed. It is currently not expected until 2026. According to reports, Apple's own AI model does not meet the expectations in Cupertino. However, this has not yet been rejected.

Controversial internally

The idea of using an external model is said to be controversial internally. The company's own AI team is demoralized, writes Bloomberg. There was a threat of emigration, especially as competitor Meta was currently poaching talent. However, Apple wants to stick to using its own models for processing on the devices anyway. The Foundation Models were recently opened up to app developers for the first time. In the fall, they will become generally usable in apps with iOS 26 and the other new operating system versions.

As part of the changes, Apple is also said to have scrapped the Swift Assist programming tool, which was also announced in 2024, in its originally planned form. When it was first presented, there was talk of a separate AI model that would be available to developers as a programming chatbot. This year, Apple explained that it had "expanded its vision". In future, ChatGPT or other models can be used in the Xcode programming environment. It is said that the original plans will no longer be pursued.

Apple, Anthropic and OpenAI did not comment on the report to Bloomberg.

(mki)