Apple has unveiled a new approach to enhance its artificial intelligence models while maintaining user privacy. In a recent blog post, the company announced that it will utilize a method that allows devices to compare synthetic datasets with samples of recent user emails or messages, but only from those who have opted into its Device Analytics program. This innovative technique ensures that user data remains on the device, with only signals indicating the closest synthetic inputs being sent back to Apple. The company aims to use these inputs to refine its AI outputs, such as email summaries. Apple has faced challenges with its AI features, leading to delays in launches and changes in leadership within its Siri team. The new AI training system is set to be introduced in beta versions of its operating systems, including iOS and iPadOS eighteen point five and macOS fifteen point five.
Why do we care?
Apple has a distinctly different approach to AI. Apple is betting on training AI using synthetic datasets matched against local user data—without that data ever leaving the device. It’s a privacy-preserving method that aligns well with Apple’s brand and regulatory climate (think GDPR, CPRA, and upcoming AI legislation). But the technical compromise is real:
Apple is still lagging Google and Microsoft in real-world AI utility. Gemini and Copilot already ship with real-time agents; Apple is still in beta and using synthetic data.
That said, many provides I speak to are still struggling to find real world utility in Copilot – and that might give Apple a chance to catch up. There might not be true first mover advantage here.