Apple’s AI division is in flux. A wave of high-profile exits has raised questions about how the company is balancing innovation with its long-standing focus on privacy. At the center of it all is a choice Apple made: to keep its AI models closed rather than open source them. That decision, insiders say, frustrated engineers and may have triggered the recent talent exodus.
The company’s top AI researchers reportedly pushed to release some of their models as open-source software. The goal was twofold: gain credibility in the AI research community and invite outside experts to help improve the technology. But the move would have also exposed a problem Apple wasn’t ready to explain publicly. Its models don’t perform nearly as well when optimized to run directly on iPhones instead of cloud servers.
Why Open-Sourcing Was Rejected
According to The Information, Craig Federighi, Apple’s senior vice president of software engineering, shut down the proposal. In an internal email to Ruoming Pang, who leads the foundation models team, Craig Federighi pushed back on the open-source idea. He said the market already had enough open models. He also worried Apple’s versions would reveal just how much performance had been lost when optimized for phones instead of high-powered machines.
That’s the tradeoff Apple made with its “on-device first” strategy, a core part of its Apple Intelligence rollout. By avoiding the cloud, Apple protects user privacy. But this approach has serious technical limitations. Some of the company’s best researchers are now leaving because of it.
Confusion, Discontent, and External Options
Another factor driving the unrest is internal miscommunication. According to the same report, researchers working on Siri were blindsided by delays in feature development. Until then, they had received only positive feedback. The shift came without warning.
Meanwhile, Apple is reportedly in talks with OpenAI, Anthropic, and Google to use their large language models. That alone signal something clear: Apple may not fully trust its in-housework to power the next version of Siri.
The Information also reports that Apple is now reevaluating compensation packages to retain the remaining members of its foundation models team.
This internal push and pull between privacy and performance, between control and collaboration, has created deep divisions within Apple’s AI leadership. What happens next depends not just on technology but on whether Apple can hold onto the people building it.
