
Apple is exploring the idea of utilizing AI technologies from Anthropic PBC or OpenAI to revamp Siri, moving away from its existing models. This could mark a significant shift in the company’s approach as it seeks to enhance its AI capabilities.
According to sources, Apple has engaged in discussions with both companies regarding the implementation of their models for Siri. They have requested customized versions that can operate on Apple’s cloud infrastructure for testing purposes.
If Apple follows through with this plan, it would signal a major change. Currently, the company relies on its own AI technologies, known as Apple Foundation Models, and had intended to update Siri using these models by 2026.
Switching to Anthropic’s Claude or OpenAI’s ChatGPT for Siri would indicate that Apple is facing challenges in keeping up with advancements in generative AI—the most significant technological trend in years. While Siri can already utilize ChatGPT for internet-based queries, it primarily runs on Apple’s technology.
Apple’s efforts to explore third-party models are still in the preliminary phase, and no final decision has been made. A project known internally as LLM Siri is still under development using Apple’s models.
Implementing this change—possibly as soon as next year—could enable Apple to provide Siri features comparable to those of AI assistants found on Android devices, potentially improving its reputation in AI.
Representatives from Apple, Anthropic, and OpenAI have not commented on the situation. Following this news, Apple’s stock experienced a rise of over 2%.
Siri’s Challenges
The initiative to assess external models was launched by Mike Rockwell, Siri’s lead, and Craig Federighi, who heads software engineering. They took over Siri’s management after John Giannandrea, Apple’s AI leader, was removed amid concerns about Siri and Apple’s AI performance.
Since starting in March, Rockwell has directed his team to evaluate whether Siri would perform better with Apple’s AI models or with third-party solutions such as Claude, ChatGPT, or Google’s Gemini.
After extensive testing, Rockwell and his colleagues determined that Anthropic’s technology showed the most potential for improving Siri, prompting discussions with Anthropic about deploying Claude.
Originally launched in 2011, Siri has not kept pace with popular AI chatbots, and Apple’s efforts to update its functionalities have faced numerous setbacks.
Last year, Apple announced new Siri features meant to leverage user data and analyze on-screen information to improve responses. However, these advancements have been consistently delayed and are now slated for release next spring instead of the initially proposed early 2025.
Uncertainty in AI Development
Inside Apple, the AI team is working under significant uncertainty, with executives still evaluating various paths forward. A significant budget has already been set for 2026 to support Apple’s own models, but future plans remain unclear.
Still, leaders like Federighi and Rockwell have started to recognize that embracing external technology might lead to a quicker turnaround. They no longer feel the need to rely solely on their internal models, which they currently consider less effective.
Licensing technology from outside sources resembles a strategy used by Samsung Electronics, which brands its features under Galaxy AI while utilizing Google’s Gemini. Similarly, Anthropic’s technology is currently used by Amazon for its Alexa service.
Moving forward, if Apple’s technology improves, executives believe it’s vital for the company to retain ownership of AI models as they become increasingly critical for product functionality. Apple is also working on various projects, including a tabletop robot and AI-driven glasses.
Additionally, Apple has considered acquiring Perplexity to strengthen its AI efforts. There were also short negotiations with Thinking Machines Lab, a startup founded by former OpenAI CTO Mira Murati.
Morale Issues within the Team
Apple’s AI models are developed by a dedicated team of around 100, led by Ruoming Pang, a distinguished engineer who transitioned from Google in 2021. He reports to Daphne Luong, a senior director overseeing AI research.
Luong was a close associate of Giannandrea, whose influence at Apple has been waning. Federighi and Rockwell have taken on more responsibility in this area as well.
The prospect of adopting third-party technology has affected team morale. Some members feel disheartened by Apple’s consideration of external solutions, which suggests that the responsibility for the company’s AI shortcomings may fall on them. Many are exploring lucrative offers from companies like Meta Platforms, which is known to pay substantially more than Apple.
Some engineers at Meta are reportedly being offered annual packages ranging from $10 million to $40 million to join their Superintelligence Labs. In contrast, Apple typically pays its AI engineers significantly less.
Recently, a senior researcher in large language models left Apple after nearly eight years, raising concerns about the difficulty of finding a replacement given the competitive job market and higher salaries offered by other firms.
Further, Apple nearly lost its team behind MLX, a key open-source platform for building machine learning models on Apple silicon. After threats to leave, Apple offered counter-proposals to retain the talent.
Negotiations with Anthropic and OpenAI
During talks with both Anthropic and OpenAI, Apple has requested customized versions of Claude and ChatGPT that can function on its Private Cloud Compute servers. This setup, based on high-performance Mac chips, aims to enhance user privacy.
Currently, some AI features run on users’ devices instead of cloud servers, which are slower and less powerful. These models are used for tasks like summarizing emails and creating custom emojis.
Apple plans to allow third-party developers access to its on-device AI models later this year, enabling app developers to integrate AI features into their applications.
However, there are no announced plans to provide access to the cloud models yet since existing infrastructure may not support a sudden influx of new third-party features.
Although Apple is not planning to move away from its in-house models for on-device or developer applications, there are concerns among the foundational models team that shifting to third-party solutions for Siri might lead to broader changes in the future.
Last year, OpenAI expressed interest in developing on-device models for Apple, but the company showed no interest at that time.
Since December 2024, Apple has been utilizing OpenAI for certain features. Besides responding to general knowledge inquiries in Siri, ChatGPT can assist in the Writing Tools feature. An upcoming iOS update will introduce a ChatGPT option for generating and analyzing images.
Apple and Anthropic have reportedly struggled to finalize financial terms during their negotiations. Anthropic is seeking a multibillion-dollar annual fee that escalates over time, which has led Apple to consider engaging with OpenAI or other companies if it decides to proceed with employing third-party technology.
Shifts in Management
If Apple manages to finalize a partnership, it may further diminish Giannandrea’s influence, who joined Apple in 2018 and advocates for in-house development of large language models.
In addition to losing Siri oversight, Giannandrea has been stripped of responsibilities over Apple’s robotics division. Other important teams, like Core ML and App Intents—responsible for integrating AI into apps—were reassigned to Federighi’s software engineering group.
Apple’s foundation models team had been working on creating large language models to assist employees and developers in coding with Xcode, but that project was canceled about a month ago. Instead, Apple will launch a new version of Xcode later this year that allows developers to access third-party programming models, with options including ChatGPT and Claude.
If you would like to see similar Tech posts like this, click here & share this article with your friends!