Apple weighs abandoning its own AI for Siri as it tests models from OpenAI and Anthropic
Key Points
- Apple is considering operating Siri with AI models from Anthropic or OpenAI in future instead of using its own models.
- The possible change in strategy is causing internal unrest and threats of departure in the AI team. In-house projects such as Swift Assist have been discontinued and Xcode is to use external models.
- Anthropic is considered the favorite, but high costs could drive Apple to OpenAI. Small AI functions remain local on the devices.
Apple is considering whether to have Siri run on AI models from Anthropic or OpenAI, Bloomberg reports. Such a move would mark a significant shift for the company and signal that Apple's own AI development has not kept pace.
Discussions are ongoing with both Anthropic and OpenAI to provide customized versions of their models for Apple's own cloud servers. Up to now, Apple has relied on its own foundation models and planned a new Siri version based on them for 2026. Adopting external models would be a sign that Apple can no longer compete at the top of the generative AI field.
The push for this transition comes from Siri lead Mike Rockwell and software chief Craig Federighi, who took over Siri after former AI chief John Giannandrea was moved aside. After testing Anthropic's Claude, OpenAI's ChatGPT, and Google's Gemini, Apple's team found Claude performed best. However, Anthropic is demanding a multi-billion dollar annual fee, which could ultimately drive Apple toward OpenAI.
Apple could rely on external models on internal infrastructure
The potential shift has stirred unrest inside Apple. Members of the foundation models team view it as a lack of confidence in their work and are considering leaving for higher-paying competitors like Meta or OpenAI. Leading LLM expert Tom Gunter recently left the company, and the MLX team, which manages Apple's open-source AI framework, threatened to quit before management intervened.
If Apple decides to go with an external model, it still plans to run it on its own servers equipped with M-series chips, not on third-party infrastructure. According to Bloomberg, this is intended to maintain privacy and control. Preliminary testing is already underway. Smaller AI features, such as Genmojis and email summaries, would continue to run directly on Apple's devices.
Meanwhile, Apple is scaling back other AI projects. The Swift Assist coding model has been scrapped, with future versions of Xcode expected to integrate ChatGPT or Claude. Giannandrea has also lost oversight of robotics and the Core ML teams. Apple reportedly considered acquiring AI startup Perplexity, but no agreement has been reached.
AI News Without the Hype – Curated by Humans
As a THE DECODER subscriber, you get ad-free reading, our weekly AI newsletter, the exclusive "AI Radar" Frontier Report 6× per year, access to comments, and our complete archive.
Subscribe now