Apple Considers External AI Models for Siri Amid Development Challenges

Apple is reportedly considering a significant shift in its AI strategy, exploring the possibility of integrating external language models from Anthropic or OpenAI to support its virtual assistant, Siri. This move comes as the tech giant faces challenges in developing its own AI models, known as Apple Foundation Models, which were initially planned to power a new version of Siri by 2026.

According to sources, Apple has been in discussions with both Anthropic and OpenAI, evaluating the potential of their models to enhance Siri's capabilities. The company has requested that these firms train customized versions of their models to run on Apple's proprietary cloud infrastructure. This approach aims to ensure that user privacy is protected by leveraging Apple's own servers, which are based on high-end Mac chips and currently used for running complex, in-house AI models.

Internal testing has shown that Anthropic's Claude technology is particularly well-suited to meet Siri's requirements. This assessment has led to preliminary negotiations between Apple and Anthropic, with the tech giant also maintaining contact with OpenAI to keep its options open. High-ranking executives, including the heads of the Siri team and software engineering, have expressed a preference for collaborating with external providers to quickly enhance Siri's functionality, rather than relying on Apple's less competitive in-house technology.

However, the negotiations with Anthropic have hit a snag due to financial disagreements. Anthropic is reportedly seeking annual fees in the tens of billions of dollars, with significant annual increases. This has prompted Apple to consider alternative partnerships, including with OpenAI. Currently, Apple's AI functions, known as Apple Intelligence, operate on a dual-model architecture, with device-side models handling lighter tasks and cloud-side models managing more complex operations. Apple plans to open its device-side models to developers later this year, but cloud-side models remain constrained by server capacity.

Despite these discussions, Apple has not entirely abandoned its in-house AI models. The company has allocated a budget of tens of billions of dollars for the operation of its cloud-based, self-developed models by 2026. However, there are concerns within the team that a shift to external technology for Siri could signal similar changes for other AI-driven features in the future.

Siri, which was launched in 2011, has fallen behind in the competitive landscape of AI chatbots. Recent announcements of new features, such as accessing personal data and analyzing screen content, have been delayed indefinitely. The AI team at Apple is currently in a state of high uncertainty, with leadership weighing multiple strategic paths. The potential shift to external AI technology has had a significant impact on the morale of the company's AI team, with some members expressing dissatisfaction and considering offers from competitors. The departure of key personnel, including a senior language model researcher and the near-exodus of the MLX open-source system team, has further exacerbated the talent drain at Apple.