Thought Leadership
Apple wants to improve its AI functions by analyzing real user data directly on the devices – without sending it to servers. The aim is to increase the quality of Apple Intelligence while at the same time protecting user privacy.
Synthetic data reaches its limits
Until now, Apple has mainly relied on synthetic data, i.e. artificially generated content, to train AI models. However, this does not always accurately reflect real user input, which can lead to inaccurate results – for example with text summaries or notifications.
New approach: Local data comparison
The new approach compares synthetic content with real emails stored locally on the device to improve the artificial data. “Our goal in creating synthetic data is to generate sentences or emails that are similar enough in topic or style to real content to improve our models for summaries – without Apple collecting emails from the device,” Apple said in the blog.
The content never leaves the device and is only used for comparison, not for training the models.
Focus on data protection remains
Apple also relies on Differential Privacy to recognize frequent inputs without making individuals identifiable. The system “recognizes popular inputs and patterns while mathematically guaranteeing that unique or rare inputs cannot be traced”, according to the company.
Only for participants in the analysis program
The new functions only apply to users who expressly agree to the analysis in the device settings.
Rollout in upcoming updates
The function will be introduced in beta versions of iOS and iPadOS 18.5 as well as macOS 15.5. Major Siri improvements are planned for 2026.
